Changelog

What's new in Saka Panel. 147 releases so far. Agent updates roll out automatically and are signed, so your servers stay current without you doing anything. Want something that isn't here? Request a feature.

Nov 202026
feature1 updates#

Endpoint in the web panel

A new Endpoint page in the panel: list your Endpoints, set one up on a server, and see the first admin password once.

Apps

  • FeatureThe panel now has an <b>Endpoint</b> menu. Pick a connected server with a public IP, optionally name it and set the login page title, and Saka installs MeshCentral for you (about 5 to 10 minutes) while the page shows each step. The initial admin password appears once in a dialog and is not stored. Each Endpoint card links to the address and to the site page for backups and releases. The list is also available at <code>GET /api/v1/endpoint</code>.
Nov 192026
feature1 updates#

Saka Endpoint: managed remote desktop on your own server

Install MeshCentral as a service on your own server in one step: remote screen, terminal and files, with no per-device fee.

Apps

  • Feature<b>Saka Endpoint</b> installs MeshCentral (open source remote desktop and device management) on a server you own, as a binary service with automatic HTTPS. Install it with the <code>endpoint_pasang</code> MCP tool or <code>POST /api/v1/endpoint</code>. Saka creates the service site, saves the settings, downloads the package (pinned version and sha256), creates the first admin before the service opens, and removes the initial password from the server afterwards. The initial admin password is returned once. Sign-up is off, 2FA is required, MeshCentral listens only on the loopback interface behind Saka's web server, and the daily backup holds the database and agent keys, so existing agents reconnect after a restore. Tested end to end on real VMs, including a Linux agent connecting through the proxy and reconnecting after restoring from backup.
Nov 192026
feature1 updates#

Binary services are now backed up and can be restored

The data folder and environment variables of a binary service are included in daily backups, S3 upload, restore and the restore test.

Apps

  • FeatureBinary services (the <code>layanan</code> site type) were skipped by backups. Their data folder and environment file now go into the same backup archive as other sites: automatic daily backup (keeps 7), manual backup, upload to your S3 bucket, <b>Restore</b> with an automatic safety backup of the current state, and the monthly restore test. The binary and its releases are not included (the last five stay on the server). A service can list folders to leave out (<code>cadangan_kecuali</code>), for caches it rebuilds by itself. The copy is taken while the service runs. Restoring stops the service, replaces the data folder as a whole, rewrites the environment file, starts the service again and waits until it is healthy.
Nov 152026
feature1 updates#

Domain names in the egress lists

The seen and blocked lists now show the domain name the app asked for, not just the IP.

Apps

  • FeatureSaka passively reads the DNS replies a container receives and remembers which name each IP answered. The monitor, allowed and blocked lists show that name next to the IP, with an Allow domain button that adds it to the allow list (safer than an IP for services behind a CDN). Nothing is changed or stored beyond name and IP; DNS over TCP, DoH or DoT is not visible, so those stay IP only.
Nov 152026
feature2 updates#

Egress allow-list presets

One click adds the domains of common services to an app's egress allow list.

Apps

  • FeatureThe Restrict card now has presets: Telegram, WhatsApp, Discord, Slack, OpenAI, Anthropic, Google Gemini, OpenRouter, DeepSeek, Groq, Mistral, xAI, Together, Perplexity, GitHub, and npm and pip. Click one to add its domains; click again is a no-op. Every domain was checked to resolve.
  • FeatureAI and API: aplikasi_egress_preset lists them, and aplikasi_egress_atur (and PUT .../egress/batas) accept a preset list of ids. Note: wildcards are not supported, so WhatsApp media (rotating CDN names) needs names added from what Simulate records.
Nov 152026
feature3 updates#

Restrict where an app may send data (egress limit)

Allow list for the internet addresses an app container may contact, with a safe simulate mode first.

Apps

  • FeatureNew Restrict card for OpenClaw, Hermes and other apps. Three modes: Monitor (default, no rules), Simulate (the app keeps working and Saka records what WOULD be blocked) and Restrict (internet connections outside your allow list are refused at once). The allow list takes domains, IP addresses and IP blocks, with optional port. Addresses already seen in the monitor log and blocked ones have an Allow button.
  • FeatureSafety: private addresses, replies to incoming connections and the server DNS are always allowed. Saka tests from inside the container network that the rules really apply, and cancels Restrict by itself if not. A 10 minute trial switches back to Simulate unless you confirm. Rules come back by themselves if the firewall is flushed or the container is recreated.
  • FeatureAI and API: aplikasi_egress_batas (read) and aplikasi_egress_atur (change, on the Telegram approval list). IPv6 is not covered yet; the Restrict mode is off for every app until you turn it on.
Nov 152026
feature1 updates#

Database proxy holds new connections during a failover

New connections wait for the new primary instead of failing during the switch.

Databases

  • FeatureThe database proxy on your app server now holds new connections (up to about 15 seconds) while the primary switches, instead of refusing them. In a test with a 9 second gap, no new connection failed; before, 38 of 106 attempts failed. Open connections still drop at failover, so keep retrying in-flight transactions. Applies when the proxy is installed or reinstalled.
Nov 162026
fix1 updates#

External checks, tunnels and load balancers keep their uptime history

Their status and daily uptime were wiped every time the control plane started and every 6 hours; they are now kept.

Monitoring

  • FixA cleanup job that is meant to drop the uptime data of deleted sites also removed the status and daily uptime of external checks, tunnels, load balancers and Kubernetes hosts, because they share the same table. It ran whenever the control plane started and every 6 hours, so the 30 day uptime of those could never grow beyond a few hours and a check that was already down lost its was-up state (the alert for going down waits until it has been up once). The job now only removes data of sites that no longer exist.
Nov 162026
improvement1 updates#

Projects can include external checks

External checks (uptime from the internet) can now be members of a project, so the project page shows whether the public address answers.

Projects

  • ImprovementA project can now include external checks next to its servers, databases and load balancers. The project page shows each check with whether it answers, the HTTP code, the response time and the 30 day uptime, and lists checks first because they show what users see. Add them from Add service, or with the MCP tool proyek_atur using the kind pantau.
Nov 162026
fix1 updates#

Database nodes stop accepting connections the moment the server shuts down

A database node that is being shut down now refuses new connections at once, so the proxy moves on instead of passing the error to the app.

Databases

  • FixWhen a server hosting a database node was shut down or rebooted in a controlled way, PostgreSQL kept answering for about 10 seconds with the error that the database system is shutting down, and a proxy could pass that error to the app. The agent now closes the database port the moment the system starts shutting down, so connections are refused and the proxy retries on another node. In a real failover test the failures dropped from 57 of 572 new connections to a few. A normal agent restart (for example an agent update) does not close the port. Applies to every database cluster as soon as its servers update the agent; the proxies themselves need no reinstall.
Nov 162026
feature1 updates#

A machine-readable list of the API routes

GET /api/v1/rute and the MCP tool rute_api list every REST route, so AI agents no longer have to read the code.

AI and MCP

  • FeatureGET /api/v1/rute (and the MCP tool rute_api) lists every REST route with its method, path, path parameters, group and access kind (public, login or admin). Filter with the group or a piece of the path. Admin routes only show to admins. The list is generated from the code and guarded by a test so it cannot fall behind; it does not include request body schemas yet.
Nov 162026
improvement1 updates#

Load balancers: copy the CNAME and see when it is in place

DNS-mode load balancers get a copy button for the CNAME target and a check that watches your domain until it points there.

Load balancers

  • ImprovementOn a DNS-mode load balancer the CNAME target now has a copy button, and a new check asks whether your domain already points to it: in place, wrong target (including a CDN proxy) or not found yet. It re-checks every 10 seconds for up to 10 minutes while DNS propagates. Also available through the API (GET /api/v1/lb/{id}/cname) and the MCP tool lb_cname_cek.
Nov 162026
improvement1 updates#

New servers: connected no longer means ready

A server from a cloud account now says it is still preparing itself while cloud-init or the first security updates hold the package lock.

Servers

  • ImprovementA server created through a cloud account used to show connected as soon as the agent registered, while cloud-init and Ubuntu's first automatic updates still held the apt lock for a few minutes, so your own apt commands failed. Saka now reports why a new server is not ready yet (cloud-init or initial updates, only in its first hour) and the dialog and the cloud server list (siap_pakai, menyiapkan) show it until it is done. Saka's own installs already waited for that lock up to 10 minutes.
Nov 162026
feature1 updates#

Binary services: limit which internet addresses a service may contact

Binary (non-Docker) services get the same egress limit as apps: simulate, then restrict.

Sites

  • FeatureA binary service (site type service) now has the card Restrict the addresses it may contact. Saka recognizes the service by its own system account and adds firewall rules on the outgoing side. Simulate records the internet destinations that would be blocked without rejecting anything; Restrict rejects what is not in the allow list, after Saka checks the enforcement as the service account. Private addresses, loopback, replies to incoming requests and DNS stay allowed. Also available through REST (GET/PUT /api/v1/situs/{id}/layanan/egress) and MCP (layanan_egress_batas, layanan_egress_atur). Domain names do not appear in the destination list and IPv6 is not covered.
Nov 162026
improvement1 updates#

Database proxy: a badge shows when an installed proxy is out of date

The cluster page flags database proxies installed before the latest fixes and offers Update.

Databases

  • ImprovementEach database proxy on a server now reports which configuration revision it runs. The cluster page shows an old version badge next to proxies that predate the latest fixes (holding connections during a failover and the PostgreSQL health check) and highlights the Update button. Updating drops database connections open on that server for a few seconds.
Nov 162026
improvement1 updates#

Kubernetes: alerts for a slow etcd disk and k3s restarting on its own

Saka now tells you when a control plane disk is so slow that etcd may lose its leader, and when k3s restarts by itself.

Kubernetes

  • ImprovementSaka now watches two more signals on the control plane and sends a Telegram alert: etcd disk writes slower than 8 seconds in the last 15 minutes (close to the point where etcd loses its leader), and the k3s service restarting on its own without a server reboot. Both close with a recovery message. On 15 Nov 2026 a slow disk made etcd lose its leader, k3s restarted and pods were recreated; this is how you would have heard about it at the time.
Nov 152026
improvement1 updates#

Database proxy also handles a primary that is shutting down

Fewer failed connections when the PostgreSQL primary server is stopped or restarted.

Databases

  • ImprovementThe PostgreSQL proxy on your app server now checks that each node really answers the database protocol, and it holds new connections for up to about 30 seconds (it was 15). In a real failover test (primary server shut down), 2 of 233 new connections failed and one waited 28 seconds and succeeded; the previous version failed 53 of 570. Applies when the proxy is installed or reinstalled; new clusters also rotate the PostgreSQL log (20 MB, 3 files).
Nov 152026
feature4 updates#

Binary services: run a non-Docker app with a domain and HTTPS

Run a single binary (for example a Go app) under systemd, with automatic HTTPS, releases and rollback.

Sites

  • FeatureNew site type, Binary service. Saka runs your binary under systemd as its own locked-down account, serves your domain with automatic HTTPS, keeps your environment variables on the server only, and checks health on every release.
  • FeatureInstall a release from a file already on the server (a binary or a .tar.gz) or from an https URL with its SHA-256. If the new release is not healthy in time, Saka returns to the previous one by itself and shows the log. The last 5 releases are kept for one-click rollback. A pre-command (for example a migration) runs once before the switch.
  • FeaturePanel: Create site, Binary service, and a Service card (state, releases, variables, settings). AI and API: layanan_status, layanan_rilis, layanan_rollback, layanan_env, layanan_atur, layanan_kontrol.
  • FeatureNot in this first version: Saka backups of the data folder, moving to another server, and two-location replicas. There is a few seconds of downtime while a release restarts.
Nov 152026
feature4 updates#

Cluster recovery, import owner, bigger VMs and SSH key at creation

Four fixes that came from a team running a real cluster on Saka.

Databases

  • FeatureNew Recover node action (and MCP tool database_pulihkan_node): starts cluster containers that were stopped on one node, in order (etcd, database, proxy). Crashes and reboots were already recovered by Docker; this covers a manual docker stop or a failover drill. Cluster diagnosis now points to it.
  • FeaturePostgreSQL import accepts an owner (an app account you created in the cluster). The database and every schema, table, sequence, view, function and type inside it are owned by that account, so the app no longer needs the superuser. Objects that belong to extensions are left alone.

Servers

  • FeatureCloud accounts: a new xbesar size (16 GB RAM) for databases, and on IDCloudHost a custom size such as kustom-4-16-100 (4 vCPU, 16 GB RAM, 100 GB disk).
  • FeatureCreating a server from a cloud account can include your public SSH key (kunci_ssh), added so you can log in to the new server.
Nov 152026
feature2 updates#

Egress list and test sandbox now in the panel

Two new cards so you do not need an AI tool to use them.

Apps and sites

  • FeatureServer, app page: a new card lists the internet addresses the app has contacted (OpenClaw Advanced tab, and at the bottom of Hermes). It only observes, nothing is blocked.
  • FeatureSite page: a new Test sandbox card runs one command on a throwaway copy of a WordPress, PHP or static site (no internet) and shows the output and which files changed. The live site is not touched.
Nov 152026
feature1 updates#

See where your AI assistant connects on the internet

A new read-only view lists the outside addresses that OpenClaw, Hermes and other apps contact from their containers.

Apps

  • FeatureNew MCP tool and API call: app egress. For each app it lists the internet destinations its container connected to (address, port, first and last seen). It only observes and blocks nothing, so it cannot break a running bot. Use it to see where an assistant sends data before deciding to restrict it. It reads the kernel connection table, so nothing is installed and no firewall rule changes.
Nov 152026
feature1 updates#

Test sandbox for sites

Run a command on a throwaway copy of a site to scan it or try a change, with no risk to the live site.

Sites

  • FeatureNew MCP tool and API call: test sandbox. It runs one command on a copy of a WordPress, PHP or static site in a container with no internet, a read-only root, and limits on memory, CPU and processes, then shows the output and which files would change. The live site is never touched.
Nov 152026
feature2 updates#

Permission-scoped tokens now work through MCP

An AI agent with a token limited to sites can deploy and edit sites, and nothing else.

Team and access

  • FeatureMCP now follows the permissions of a scoped token. The AI sees only the tools for the areas you ticked (sites, servers, databases, kubernetes, networking, monitoring) plus all read-only tools; other tools are hidden and refused with a message saying which permission is missing. Tools that delete need the delete permission. Risky actions still ask for your approval on Telegram. Before this, a scoped token was read-only through MCP.
  • FeatureMenu: Projects moved from the top bar into the account menu, above Settings.
Nov 142026
feature3 updates#

Custom team roles and permission-scoped API tokens

Build your own team roles from seven permissions, and create API tokens that may only change certain areas.

Team and access

  • FeatureSettings, Team now has Custom roles. Pick from sites, servers, databases, kubernetes, networking, monitoring and delete, for example a role that can deploy sites but cannot delete. Viewing is always allowed; changing needs the permission for that area, and deleting also needs delete. A role takes effect immediately when edited and cannot be deleted while members use it. Custom roles can never manage the team or the owner's personal settings.
  • FeatureAPI tokens can now be created with selected permissions, for example a CI token that may only change sites. A scoped token cannot manage other tokens, the team, cloud accounts or the terminal, and through MCP it is read-only. Existing tokens are unchanged.
  • FixTelegram alert buttons such as Restart now follow the member's permissions instead of only checking for the Viewer role.
Nov 112026
fix2 updates#

Rebooting a node that holds a single-node volume, and a calmer uptime monitor

The reboot guide now handles apps pinned to one node by their volume, and a few dead addresses no longer slow down every monitoring round.

Kubernetes

  • FixAn app whose pods all share one single-node (RWO) volume cannot be moved off a node while it keeps running: the replacement pod waits for the volume and the disruption budget blocks the last pod, so the drain stalled and the reboot was cancelled. The reboot guide now detects this before changing anything, explains it, and offers a short pause (about 1 to 2 minutes). On your approval Saka stops that app, empties the node, starts the app again on another node, then reboots. The original replica count is stored on the app itself, and Saka restores it automatically if it restarts mid-way.

Monitoring

  • FixAddresses that fail again and again (a customer site that is simply down) made each monitoring round wait for them and then confirm from other locations one by one, which could trigger a false capacity alert. Other locations are now asked at the same time, and an address that has failed for 15 minutes in a row is checked every 5 minutes in the background, outside the timing of the main round. Down alerts are unchanged; the all clear for such an address can arrive up to 5 minutes later.
Nov 92026
fix1 updates#

Update system now installs the new kernel on Ubuntu and Debian

On Ubuntu and Debian, Update system reported everything up to date while the fixed kernel was never installed, so a reboot did not close the flaw. It now installs the new kernel.

Servers

  • FixUpdate system used a plain upgrade, which never installs new packages, and a new kernel is a new package. The kernel meta packages were held back and the kernel check kept offering the same button. Update system now also upgrades the installed kernel meta packages (generic, aws, azure, gcp and so on), which installs the new kernel. Nothing is removed and no other kernel flavour is added. After it finishes the server shows a pending reboot.
Nov 92026
fix3 updates#

Safer node reboots for Kubernetes clusters

Kubernetes nodes are rebooted one at a time with a guide on the cluster page; Saka checks the cluster first and waits for storage to heal afterwards.

Kubernetes

  • Checks before a rebootSaka now refuses to reboot a node while another node is not Ready, while the control plane would lose etcd quorum, or while a Longhorn volume is still copying data. Nothing is changed when it refuses, and the message says what to wait for.
  • Waits for healthy dataAfter the node is back, the cluster stays marked as busy until every Longhorn volume has all its copies again, so the next node cannot be rebooted too early. The Telegram message says when it is safe to continue.
  • Step-by-step guideThe cluster page now has a security update guide: nodes are listed workers first and control nodes last, only the node whose turn it is has an active button (update the system, then reboot), and the next one unlocks when the previous node is back and its data is healthy. The patch card links to it.
Nov 12026
feature1 updates#

CloudLinux 8, 9 and 10 can now connect

Servers running CloudLinux 8, 9 or 10 can be connected like AlmaLinux and Rocky Linux.

Servers

  • CloudLinuxThe install command no longer refuses CloudLinux. It is treated like the other RHEL-family systems; it has not been through the full end-to-end test yet, so tell us if anything misbehaves.
Nov 12026
feature1 updates#

Server limit per account, set by admins

Admins can now set how many servers an account may connect, as a default for everyone and as an exception for single accounts.

Accounts

  • Server limitThe Users tab in the admin area has a default limit for all accounts (10 unless changed) and a Server limit column to give one account its own number. Leave it empty to follow the default. The limit applies when installing the agent, adding a server over SSH, and creating a cloud server.
Nov 12026
feature5 updates#

Observability: slow queries, event markers, longer history, habit-based alerts

Observability now shows slow MySQL queries and PHP scripts, marks reboots and updates on the charts, reads history beyond 7 days from your S3 archive, alerts on unusual spikes, and keeps saved log searches.

Add-ons

  • Slow tabReads the MySQL/MariaDB slow query log and the PHP-FPM slowlog that are already on the server, groups queries by shape (values replaced with ?), and lists the slowest scripts. Saka changes no database or PHP setting; when a log is off it shows how to turn it on.
  • Event markersReboots, agent updates, system package changes, PHP file changes, and alerts that fired appear as dashed lines on the charts and in a list below them, so you can see what changed right before a spike.
  • History beyond 7 daysWith the S3 archive on, the Overview can show 14 and 31 days, read from encrypted daily summaries (5-minute averages) in your own bucket.
  • Habit-based alertsNew alert rules compare CPU, TCP connections, or network traffic with the same hour on previous days, with a minimum value, so a normally busy server stays quiet and an odd spike does not.
  • Saved searches and live followSave a log search under a name and reuse it, or follow results live (reloaded every 10 seconds, last hour).
Oct 312026
feature3 updates#

PHP file monitoring, and your own retention for the S3 archive

Observability can now watch PHP files for new or changed shells, and the S3 archive retention is yours to set, down to the day or never.

Add-ons

  • PHP file monitoringA new PHP files tab records PHP files (plus .htaccess and .user.ini) that are new, changed, or deleted in your web folders: Saka sites, /var/www, public_html, aaPanel, and folders you add. Each event keeps the time, size, and SHA-256 of the content, so you can match it against the access log to see where it came in from. Shell-like content, PHP in upload or cache folders, and an .htaccess that runs images as PHP go to Telegram right away; ordinary changes are notified only when few, so plugin updates do not flood you. The scan is gentle (names only, content read only for changed files) and skips framework caches.
  • Retention you chooseThe S3 archive retention is no longer fixed to 30, 90, or 365 days: pick a preset, any number of days from 1 to 3650, or never delete automatically and manage lifecycle in your own bucket. The note about raw logs is now just that, a note: the bucket is yours, so the choice is too.
  • FixLocal error-group files from the Errors tab were not being pruned after 7 days; they are now.
Oct 312026
feature3 updates#

Observability archive to your own S3, for investigating incidents

Observability data and logs can now be archived, encrypted, to your own S3 bucket for 30, 90, or 365 days, and searched later: which IP broke in, and what did it open.

Add-ons

  • S3 archiveA new S3 archive tab uses the same destination as Backup to S3 (IDCloudHost Object Storage, Amazon S3, Cloudflare R2, and others). Metric, endpoint, and error files go up once their hour has passed. Everything is encrypted with a key that only exists on your server, objects end in .enc so they never mix with site backups, and old objects are pruned after the retention you pick. The server disk still keeps 7 days.
  • Raw logs, opt inRaw logs are archived only if you choose them. The default is errors and warnings, and in that mode web access keeps only 4xx and 5xx responses; All logs is a deliberate choice with a warning about personal data and S3 cost. Sources are the system journal, SSH logins, web access of every Saka site, nginx, Apache, and hosting-panel (aaPanel, cPanel) logs plus PHP logs found on the server, and running containers, hour by hour. That covers a hack through a PHP shell on a site Saka does not host: the web server access log shows the POST that uploaded it.
  • Search the archiveSearch logs can now read the archive: tick S3 archive, pick a source (or all), a range of up to 31 days, and an end date, then look for an IP or an address. Errors accepts 30 and 90 day ranges. Web access and SSH logins are also available as live sources.
Oct 312026
feature1 updates#

Observability is now its own menu

Observability moved out of the server page into a top-level menu with a server picker, because it now has four tabs of its own.

Add-ons

  • A menu of its ownAccounts with the Observability add-on get an Observability item in the top bar: pick a server, then Overview, Errors, Search logs, and Alerts. The menu is hidden for accounts without the add-on. The server page keeps a short card that opens the menu for that server, or explains how to request the add-on.
Oct 312026
feature2 updates#

After a reboot, Saka tells you whether your apps came back

When a server restarts, the agent now waits for apps to start and then reports whether they are running again, instead of leaving you to find out later.

Servers

  • Post-reboot reportAbout 5 minutes after a server boots, Saka checks the apps it manages and the Hermes and OpenClaw installs it detected. If everything is running you get a short note; if something is not, you get the list of what is down with where to look, and a follow-up when it recovers. Agent restarts without a reboot, and profiles you parked on purpose, are not reported.

Databases

  • First steps on a new MariaDB clusterA brand-new MariaDB cluster has no database yet, so the connection address does not work until one exists. The Database users card now says so and points to Create database.
Oct 312026
feature6 updates#

Observability add-on: why a server is slow, not just that it is

A new opt-in add-on records per-minute host metrics, the processes behind every CPU spike, and slow or failing site endpoints. It is off for every account until a Saka admin turns it on, and it is free while billing does not exist yet.

Add-ons

  • Off by default, switched on per accountAdd-ons are not part of the base plan. Observability stays off, and your server records nothing, until a Saka admin enables it for your account (ask through the Enterprise button). Opening the card without it shows what it is and how to ask. The structure is per account and per add-on, ready for billing later.
  • Metrics and the process behind each spikeEvery minute the agent records CPU, I/O wait, 1/5/15 minute load, RAM, swap, network, disk read and write, and TCP connections, plus the top 5 processes by CPU and top 5 by RAM with the Docker container name. Hover over any chart to see which processes were running at that moment; it opens on the moment of highest CPU.
  • Slow and failing endpointsAccess logs of the sites Saka manages are summarized every 5 minutes per endpoint (for example GET /orders/:id): requests, 4xx, 5xx, average, estimated p95, and slowest. Sort by slowest, most 5xx, or most requested, and filter by domain. Numbers and long codes in paths become :id and the query string is dropped.
  • Light and localOne /proc read per minute, a few megabytes of text files for a week, nothing injected into your apps, and no polling from the panel. Data stays on your server for 7 days and is never stored at Saka. Turning the add-on off stops recording at once.
  • Grouped errors, log search, and alert rulesThree more tabs on the same card. Errors groups error-level lines from running containers and the system journal (numbers, IDs, and timestamps ignored, so one bug counts once) with first and last seen. Search logs reads a container or the system journal live, with keyword or regex, level, and range; no index is kept, so it costs nothing while unused. Alerts lets you set rules such as CPU, I/O wait, swap, load per core, TCP connections, one process's CPU, site p95, or site 5xx ratio held for N minutes; the agent evaluates them every minute and notifies your Telegram, with a recovery notice.
  • Honest limitsIt is not distributed tracing. Durations are server processing time, not visitor network time, p95 is the upper bound of a duration bucket, and endpoints are counted only for sites Saka manages on that server.
Oct 312026
feature12 updates#

Read-only and full-write users for a MariaDB database, patch health, and a calmer phone UI

Give an analytics team a read-only account and your app a full-write one on the same MariaDB database, see vulnerable kernel, PHP and Node.js in one card, and read the Activity log as plain sentences.

Databases

  • Database users card (MariaDB)On a MariaDB cluster, create accounts per database with Read only (SELECT and views, cannot change or delete data or list accounts) or Full write (all rights on that one database). The password is generated by Saka and shown once; New password, change access and Delete are one click, with a confirmation. Deleting a user never touches the database. Accounts apply on both data nodes at once and require SSL. Only accounts created through Saka are shown and touched; root and site accounts are not.
  • Create databaseA new cluster has no database, so the same card can now create an empty one (letters, digits, underscore, up to 64 characters; MariaDB system schemas are refused). It exists on every node immediately.

Servers

  • OpenClaw that survives a rebootA directly installed OpenClaw (systemd service, not Docker) that is not set to start at boot now gets a Readiness warning with an Enable at boot button. Saka picks the service in use (the running one, then the enabled one, then the one with log history), never touches its contents, and ignores one-shot first-boot units. The Other apps card also gets Logs, Start, Stop, and Restart for such an install.
  • Terminal sessions that surviveThe emergency terminal can keep your shell on the server (tmux, or screen if tmux is missing). Refresh the page, close the tab, or lose your PC and the commands keep running; open the terminal again with your password and the screen continues. It is on by default and can be switched off per session. Without tmux or screen the terminal still opens as before and tells you how to install one. The password is still asked every time.
  • Patch health cardOne card on the Servers page for vulnerable kernel, PHP and Node.js versions and a pending reboot. A server that only waits for the vendor's patch is a single short line. The minimum versions come from a catalogue on the control plane, so a new advisory needs no agent release. Telegram notices cover vulnerable Node.js and kernel.
  • Readiness shows problems firstChecks that are fine fold behind a button when something needs attention, and each warning's fix steps sit behind Show how, so the card is about half as tall on a phone.
  • Other firewalls recognisedA firewall owned by another hosting panel (iptables with a DROP policy) now counts as active: no false warning, and Saka does not install a second firewall on top.

Account

  • Activity log in plain sentencesRows read like "Created database user analyst" instead of action codes. On a phone each row becomes a compact card and the IP is hidden.
  • Sign in from the email buttonThe email with your sign-in code has an Open Saka Panel button that finishes signing in without typing the code (single use, expires with the code).
  • Telegram bot token errorsA wrong bot token now says what is wrong (bot name, wrong shape, revoked token, non-numeric chat ID), and an ISP block page is no longer mistaken for a bad token.

Interface

  • Dark mode contrastText on primary buttons is readable in dark mode (contrast raised from 3.2 to 6), and catalogue cards follow the theme.
  • PhoneSegmented buttons are large enough for a thumb, and tables that were wide on desktop become cards.
Oct 282026
feature5 updates#

Ready if a server dies? A readiness board and a project dashboard

One place answers whether your setup survives losing a server, and Projects now shows the real state of your database cluster and Kubernetes cluster, node by node.

Servers

  • Readiness boardA collapsible section at the bottom of the Servers page lists, per database cluster, whether a replica is streaming (and its lag), whether backups are on with no alarms, and when the backup was last proven restorable; for DNS switches the TTL and whether the standby is monitored and healthy; plus active alerts. Read-only.
  • Cluster network pre-checkPick the servers that will form a cluster and Saka checks each one (Docker, WireGuard, disk, RAM, free ports), then sends UDP and TCP packets between every pair to find provider firewalls that block the cluster tunnel, and measures round-trip time. Nothing is installed. MCP tool jaringan_periksa.

Projects

  • Application dashboardA project now opens with a one-line summary (all good, almost good, or problems) and what needs attention, linking to the source. A database cluster shows each node with its role, state and lag, and its readiness checks. A Kubernetes cluster shows every node with pods, CPU and RAM, problem pods and Saka's advice. Servers can be added too.

Security

  • Telegram approval for risky actionsOptional, off by default (Settings). When on, an API token or an AI agent cannot instantly delete a server or database, switch the primary, switch DNS, open database access or resize a VM: Saka sends you a Telegram message and runs the action only after you tap Approve (valid 15 minutes). The panel you use yourself is not affected.

Updates

  • Agent version decoupled from the control planeThe agent version is now a hash of the agent's own code, so a release that only changes the panel or control plane no longer restarts the agent on every server. Agents update only when the agent actually changes.
Oct 282026
feature6 updates#

Database clusters you can grow, shrink, diagnose and move apps onto

Add and remove data servers on a running PostgreSQL cluster, switch the primary on purpose, see why a node is unhealthy, and move an existing database onto the cluster with automatic rollback.

Databases

  • Add a data server as a replicaAdd a server to a running PostgreSQL cluster: Saka installs PostgreSQL there, joins it to the private WireGuard mesh, and copies the data from the primary while the primary and your apps keep running. If the cluster already has S3 backups, the backup configuration is copied to the new node so backups continue if it becomes primary.
  • Remove a replicaRelease a replica added this way from the cluster (its data on that server is removed; the cluster keeps running). A server that died for good can be released with a force option. The primary and the original etcd members cannot be released.
  • Make primaryPlanned switchover to another replica through Patroni, with app proxies updated first. Tucked into Advanced so it is not clicked by accident.
  • DiagnoseRead-only evidence of a node: containers, WireGuard handshakes, firewall, listening ports, recent logs without passwords, and the Patroni view, with a plain-language guess at what is wrong.
  • Replication mode and outside accessSwitch between synchronous and asynchronous replication, and allow specific IP addresses to reach the database port through the node firewall. Nothing is open to the internet by default.
  • Guided move of an existing databasePoint Saka at the old app's env file and its systemd service: Saka stops the service, copies the database in one transaction, rewrites the connection variable (the old file is kept), starts the service and checks it is healthy. If any step fails the env file is restored and the app starts again on the old database, which was only ever read. The password never passes through the panel or chat.
Oct 282026
feature7 updates#

Backups proven by restoring them, a reusable S3 destination, and one-click DNS failover

Every week Saka restores your latest database backup into a temporary copy and checks it against the live database, a failover drill measures how fast visitors move, and S3 details are entered once for all backups.

Databases

  • Automatic restore testWeekly, in the early morning on a replica when there is one: Saka restores the newest backup from S3 into a temporary copy, counts the tables in every database and compares with the live database, reports on Telegram, and deletes the copy. The live database is only read. There is also a Test restore now button.
  • Backup alarmsTelegram alerts when the first full backup does not appear, when the newest backup is older than 36 hours, or when archiving WAL to S3 is failing, once a day until fixed and again when it recovers.
  • Fix and recovery codeA Fix button moves a half-set-up backup to a fresh repository, and Show recovery code reveals the encryption passphrase (shown on demand, never stored in Saka).

Settings

  • S3 destinationEnter bucket and keys once in Settings (optional, encrypted, only with your consent) and pick them for database and site backups.
  • One-click DNS switchPoint one Cloudflare A record at a primary or a standby address and switch with one click. The token is limited to one zone, stored encrypted and never shown again.
  • Failover drillSwitches the record to the standby for real, measures how many seconds 1.1.1.1 and 8.8.8.8 take to follow, checks the standby serves, holds, switches back and measures again, then reports on Telegram. The record is always switched back, even when a step fails. MCP tool dns_failover_gladi.
  • Monitor a standby by IPAn external monitor can connect to a given IP while sending the site's own Host and TLS name, so a standby can be watched, certificate included, without changing DNS.
Oct 282026
feature3 updates#

Add a server over SSH with a key, and Bulk deploy inside Sites

Install the agent on a new server using an SSH key as well as a password, and find Bulk deploy where it belongs, under Sites.

Servers

  • Add over SSHGive Saka a server's address and either a password or a private key and the agent is installed over SSH; the credentials are used once and not stored.

Sites

  • Bulk deploy under SitesBulk deploy is now a tab of Sites instead of a separate menu.

Alerts

  • Stale traffic-spike alertsA traffic-spike alert used to stay active for days if the next day was normal; it now closes on the next check.
Oct 232026
feature4 updates#

Manage your own Hermes install, without moving it

Most Saka servers run a Hermes Agent the owner installed by hand. Saka now manages it in place: logs, restart, update, model test and downtime alerts, with nothing copied or moved.

AI agents

  • Buttons on the detected HermesOn the server page, a Hermes that was installed outside Saka now has Log, Restart, Stop or Start, Test model, Check for updates and Update Hermes. Every action runs the owner's own hermes CLI as the owner's Linux user, exactly as typing it in the terminal. Named profiles get their own buttons.
  • Downtime alertsIf a self-installed Hermes that was running stops, you get one Telegram message with Log and Restart buttons, and another when it is back. A profile you parked on purpose with hermes gateway stop is not reported.
  • Version shownThe detected Hermes shows its version; Check for updates runs hermes update --check.
  • MCP and APITool aplikasi_hermes_luar and POST /api/v1/server/{id}/luar expose the same actions to AI agents and scripts.
Oct 222026
security4 updates#

php2shell: PHP checked and Saka sites patched automatically

Two critical PHP flaws (CVE-2026-109618, CVE-2026-109619) allow remote code execution without login. Saka now checks every server for vulnerable PHP and rebuilds its own sites on the patched images.

Servers

  • PHP check in Server readinessThe agent looks for PHP on the host (including cPanel, LiteSpeed and aaPanel installs), inside every Docker container, and in Saka-managed sites, and compares versions against the patched releases (8.2.35, 8.3.36, 8.4.27, 8.5.12). Distribution backports that keep the old version number are recognised from the package changelog.
  • Telegram noticeOwners of affected servers get one message listing where the vulnerable PHP runs and what to do. A second message confirms when everything is patched.

Sites

  • Saka sites rebuilt automaticallyThree minutes after the agent starts, and daily after that, Saka pulls the latest PHP base image and recreates site containers that still run the old one. Future PHP security releases reach your sites without a Saka release.
  • Fix buttonIf an automatic rebuild failed, the readiness card offers a Fix button that runs the same rebuild on demand.
Oct 162026
fix8 updates#

Fixes from real failures this week

Fixes for problems users actually hit: SSH hardening on newer Ubuntu, moving Apache, picking an AI model, and long tasks shown as failed.

Servers

  • SSH key-only login on Ubuntu 24.04+Turning off SSH password login failed on servers where SSH starts on demand (ssh.socket). It now works there too, and the same fix covers SSH port detection for the firewall.
  • Moving Apache asideApache lines such as "Listen 80 http" or Listen in apache2.conf and conf-enabled are now moved too, so Saka can stand in front of Apache. If port 80 is still taken, the error names the program.
  • Port 80 used by a containerWhen a Docker container (for example n8n with its own Caddy) holds the web ports, the advice now explains how to move the app into Saka instead of suggesting to stop nginx/Apache.
  • Long tasksLong tasks such as taking over Hermes (multi-GB downloads) are no longer shown as failed after 20 minutes while the server is still working; the panel waits as long as the server is allowed to.

Projects

  • Empty accountAdding to a project on an account with no sites or databases yet now says so and links to where to create them, instead of claiming everything is already in the project.

Assistants

  • Hermes taken over with a local AI gatewayA Hermes that pointed at an AI gateway on the same server (127.0.0.1, e.g. 9Router) kept "typing" without answering after being taken over, because inside the container that address is the container itself. Every local address in its data (config, provider list, .env, model catalog cache) is now rewritten to the host automatically, also for existing takeovers, and you are told on Telegram.
  • Test model buttonThe Model card on a Hermes app has a Test model button: Saka runs one short chat inside the container and shows the real provider error (connection, rejected key, unknown model) without SSH.
  • Pick the model from a listFor OpenAI-compatible endpoints, OpenAI, and OpenRouter, the model field now lists the models the endpoint offers, with search. Names that differ only in spacing or dots ("claude haiku 4.5") are matched to the exact name automatically.
Oct 182026
feature2 updates#

Move a Hermes profile from a host without Docker to a real server

A Hermes installed outside Saka, including a named profile, can be moved straight to another server and taken over there, so a container-only host is no longer a dead end.

Assistants

  • Move from the detected listOn "Found on this server", a Hermes profile now has Move to another server. Saka copies the whole ~/.hermes (shared memory and other profiles included) to the target over an encrypted direct transfer, starts it there as a Saka-managed Hermes serving that profile, and only then stops the old gateway; if anything fails the old one is restarted.
  • WhySome hosts (for example systemd-nspawn or LXC containers without network rights) cannot run Docker at all. The readiness card now says so, and this gives those users a way to a managed Hermes on a normal VPS.
Oct 182026
fix3 updates#

Docker installed but not running is now detected and fixable

A server whose Docker daemon was down made installs and Hermes take-overs fail with confusing errors; Saka now checks that Docker answers, not just that it is installed.

Servers

  • Readiness cardDocker shows "installed but the daemon is not running" with a Fix button that starts it; if it still fails, the reason from the Docker log is shown.
  • Before installs and take-oversSaka starts Docker first and stops with a clear message when it cannot, instead of "cannot connect to the Docker daemon" mid-download.
  • False "Hermes already exists"With Docker down, the container check mistook the error output for an existing container and refused the take-over. Fixed (reported by a user on 18 Oct).
Oct 182026
feature3 updates#

Forensic audit: PDF report and read-only share links

Evidence bundles now include laporan.pdf, and a bundle can be shared with an auditor through a revocable read-only link, no Saka account needed.

Servers

  • laporan.pdfThe plain-language report inside the bundle is also written as a PDF, so a non-technical reader can open it directly.
  • Read-only linkShare link creates https://app.saka.work/bukti/<token>, valid 7 days: a page with the server, range, SHA-256, the agent's public key as recorded by Saka, the verification steps, and a download button. Revoke it any time; downloads through the link are logged in the action history.
  • Stuck tasks closedTasks left "running" forever when Saka restarted mid-task are now marked failed after 3 hours.
Oct 182026
feature3 updates#

Forensic audit, stage 3: signed evidence bundle

One zip per time range with the baseline, changes, timeline, raw logs, and a plain-language report; every file hashed, the hash list signed by the agent, and the zip hash recorded in Saka off the server.

Servers

  • Evidence bundleFrom the Forensic audit card pick 24 hours to 30 days and create a bundle: garis-dasar.json, perubahan.json, linimasa.json, log/ (sshd, sudo, system errors, docker events, dpkg), laporan.md for non-technical readers, SHA256SUMS, an Ed25519 signature over it, the agent's public key, and VERIFIKASI.md with the openssl commands to check it all.
  • Recorded off the serverThe zip's SHA-256 and the signing public key are stored in Saka when the bundle is created, so a bundle altered on the server afterwards no longer matches. Download from the panel; deleting removes the file but keeps the record.
  • For AI agentsserver_perbaiki forensik-bukti (dari/sampai) and forensik-bukti-daftar.
Oct 182026
feature2 updates#

Forensic audit, stage 2: event timeline

One timeline per server answers "what happened between X and Y": SSH logins, sudo, Docker, packages, boots, system errors, changes outside Saka, and Saka tasks.

Servers

  • Timeline on the Forensic audit cardPick 24 hours, 3 days, or 7 days. Sources: journal sshd (logins per user and IP, failed-login counts with top IPs), sudo commands, Docker container events (start, stop, exit code), dpkg.log or dnf history, boots, journal errors (deduplicated per unit), stage-1 changes, and Saka tasks with who ran them. Filter by source with one click. Read-only; shell history is deliberately not read (no timestamps, editable by anyone).
  • For AI agentsserver_perbaiki forensik-linimasa with dari/sampai (unix), up to 30 days.
Oct 182026
fix3 updates#

Take over an existing Hermes profile instead of installing a new one

Hermes installed with named profiles (hermes profile) is now detected per profile and can be taken over as is: memory, skills, bot, and settings come along.

Assistants

  • Per-profile detectionThe "Found on this server" card lists the default Hermes and every named profile under ~/.hermes/profiles, with its model, whether it has a Telegram bot, and which one is active. Leftover folders without an identity and deleted profiles are ignored.
  • Take over a profileChoose the profile whose bot Saka should manage. The whole ~/.hermes is copied so shared memory and other profiles are kept, the container is pointed at that profile, and the profile's own gateway service (hermes-gateway-<profile>) plus the default one are stopped; both are restored if anything fails.
  • No more duplicate installsRequested by a user whose VPS already had a custom profile: Saka used to offer a fresh install. Install is refused while a Hermes exists outside Saka; the take-over path is the way in.
Oct 182026
feature3 updates#

Forensic audit, stage 1: baseline and changes outside Saka

Saka snapshots the sensitive state of each server and tells you on Telegram when something changes that Saka did not do.

Servers

  • BaselineA few minutes after the agent starts, Saka records users and sudo rights, SSH keys per account, cron, systemd units, listening ports, SUID files, Docker containers and images, key config files (sshd, hosts, resolv.conf, passwd, shadow, ld.so.preload, rc.local, PAM) and the package list. Read-only.
  • Changes outside SakaEvery hour the server is compared with the baseline. Changes made by Saka tasks are folded into the baseline silently; everything else (you over SSH, an AI agent, an intruder) is listed on the new Forensic audit card and sent to Telegram, for example "SSH keys: added root: ssh-ed25519 ...". Package changes are recorded but not alerted.
  • Check now / Set as baselineRun a comparison on demand, and after reviewing changes make the current state the new baseline. Also for AI agents via server_perbaiki forensik-cek, forensik-periksa, forensik-dasar.
Oct 162026
feature3 updates#

Firewall controls on the server page

Turn the firewall off and on, and open or close ports, from the panel instead of SSH.

Servers

  • Open a portInstalling an app that needs its own port? Open it (e.g. 8080/tcp) from the new Firewall card; the firewall stays on.
  • Close a portPorts opened earlier can be closed again. SSH, Saka site, VPN, and platform ports are marked required and cannot be closed by accident.
  • Turn off and onThe firewall can be turned off with a confirmation (for example while installing), and turned back on with the usual port check.
Oct 162026
feature5 updates#

A working address for every new site, no domain needed

Leave the domain empty when creating a site and it opens right away at https://<name>-app.hack.id with HTTPS.

Sites

  • Automatic addressSites created without a domain get <name>-app.hack.id, pointed at the server and secured with HTTPS automatically. No DNS to set up; add your own domain later from the site page.
  • For existing sitesSites that already have a domain can add a free address from the Domain card on the site page.
  • Kept cleanSaka checks these addresses after uploads and every few hours for fake login pages and online gambling, so one bad site cannot get the whole address space blocked.
  • Follows the siteIf the site moves to another server the address follows; deleting the site or removing the address frees the name.
  • For AI agentssitus_buat accepts an empty domain, and situs_alamat gives an existing site an automatic address, so an agent can go from a folder to a public URL without asking for DNS changes.
Oct 162026
feature3 updates#

Projects: a whole app in one view

Group the sites, Git apps, databases, load balancers, and tunnels of one app into a project and see every service with its state and CPU/RAM side by side.

Projects

  • One view per appA new Projects menu groups related services, even across servers. Each service shows its type (web app, worker, cron, static site, database), domains, state, and CPU/RAM against its server's capacity. Workers, cron jobs, and app databases of Git apps are listed automatically.
  • Grouping onlyRemoving a service from a project or deleting a project never deletes or stops anything.
  • For AI agentsMCP tools proyek_daftar, proyek_status, and proyek_atur, so an agent that deploys an app can file it under its project and report on the whole thing.
Oct 162026
feature2 updates#

Deploy a folder as a site, images included, from the CLI or an AI agent

Static sites no longer need Git: upload a whole folder with one CLI command, or let an AI agent upload images and zips over MCP.

Sites

  • saka situs unggahsaka situs unggah <site> <folder> zips the folder, uploads it, and extracts it into the site (.git, node_modules, and .env files are left out), then prints the URL. Single files work too; --ke puts them in a subfolder.
  • situs_unggah for AI agentsA new MCP tool uploads binary files (images, fonts, PDFs) or a zip that is extracted in place, up to 10 MB per call. panduan_deploy now has a static-folder path without Git, so asking an agent to put a folder online ends with a working URL and images that load.
Oct 112026
feature2 updates#

Load only the MCP tools you use

AI agents can load a subset of Saka's 125 MCP tools, cutting about 25k tokens of tool descriptions from every conversation.

AI & MCP

  • Tool groupsAdd ?alat= to the MCP address, e.g. /mcp?alat=situs,pantau, to load only those groups plus the core tools. A site owner using sites and monitoring loads about 36 KB of tool definitions instead of 98 KB. Without ?alat= nothing changes.
  • In SettingsWhen creating a token, tick the groups you need and the Claude Code, OpenClaw, and Hermes commands update (the address is quoted so it works in zsh).
Oct 112026
fix2 updates#

Consistent server parameter in MCP tools

Every MCP tool that acts on a server now accepts server_id, like the other 30+ tools.

AI & MCP

  • server_id everywhereperisai_status, perisai_atur, perisai_lepas, server_cloud, server_snapshot, server_ubah_ukuran, and server_hapus_cloud used id for the server while id means a site elsewhere, which confused AI agents. They now take server_id; id still works for existing setups.

Panel

  • Shorter Telegram suggestionThe Link Telegram bar is shorter so it takes two lines on a phone. Checked on 13 pages at phone width: nothing wider than the screen.
Oct 112026
fix4 updates#

Git deploy errors that point to the real cause

Found by walking through the main flows on a fresh server like a new user would.

Apps from Git

  • App errors vs no answerIf your app starts but answers with an error (HTTP 5xx), Saka now says so instead of claiming the app does not answer and pointing at the port.
  • Missing variables namedCommon messages for a missing environment variable (Django settings, KeyError, 'X is not set', 'missing environment variable X') are recognized and the step names the variable. For Django, Saka reads your settings to find the variable the setting actually comes from (for example SECRET_KEY read from SECRET). Secrets get a hint to use Fill random.
  • Slow first requestsHealth checks wait up to 10 seconds per request (was 3), so apps with a heavy first request are not marked failed; when the port accepts connections but requests hang, the message says so and suggests a light health-check path.

Servers

  • New server clockA freshly booted server no longer shows a clock warning while its first time sync is still running.
Oct 112026
fix1 updates#

Second warning when a disk is almost completely full

You now get a clear second alert when a server's disk reaches 98%, not only the first one at 90%.

Alerts

  • Disk at 98%At 90% Saka cleans what is safe to remove and tells you once. If the disk keeps filling to 98%, you get one more message saying databases, apps, and Docker may stop when nothing more can be written, with steps to free space. Before, a disk could sit at 99% for days after a single 90% alert.
Oct 112026
fix6 updates#

Fixes from real failures: Git branches, clock check, alerts

A round of fixes chosen from the tasks that actually failed for users over the last two weeks.

Apps from Git

  • Default branch detectedWhen you leave the branch empty, Saka asks the repo for its default branch (main, master, canary, ...) instead of assuming main. Several first deploys failed only because the repo used master.
  • Repo address cleaned upPasted addresses like github.com/owner/repo/tree/dev or github.com/owner/repo/master/ are cleaned to the repo address, and a branch in the address is used when none is set (GitHub, GitLab, Bitbucket, Codeberg).
  • SSH reposIf the branch does not exist, the error now names the repo's default branch.

Servers

  • Clock checkServers whose provider blocks NTP and whose clock Saka keeps correct from HTTPS time now show the clock as OK in Readiness, instead of asking to press Fix again and again.
  • IP changesA server whose IP alternates between two addresses (two internet lines or provider NAT) now gets at most one message a day when nothing needs adjusting.

Hermes

  • Clearer model errorsEntering several models separated by commas asks for a single model, and switching provider explains that the new provider's API key is needed.
Oct 112026
fix1 updates#

Sites and apps install even where Docker Hub is blocked

If a server cannot reach Docker Hub but can reach Google's Docker Hub mirror, Saka adds the mirror to Docker's settings so every download works, including the web server and databases.

Servers

  • Docker Hub mirrorWhen Docker Hub is unreachable (blocked network or rate limit) and mirror.gcr.io is reachable, Saka adds it to registry-mirrors in /etc/docker/daemon.json, keeping your other settings, and reloads Docker without restarting containers. Servers that reach Docker Hub are not changed. Tested with Docker Hub blocked: a ready-made app installed and opened over HTTPS.
Oct 112026
fix3 updates#

More reliable app installs and Hermes takeovers

Installs no longer stall when Docker Hub cannot be reached, and taking over an existing Hermes checks disk space before touching anything.

Apps

  • Download mirrorWhen Docker Hub cannot be reached from a server (blocked network or rate limit), the agent retries the same image from Google's Docker Hub mirror (mirror.gcr.io). Download errors now always show the real cause.
  • Hermes takeoverBefore taking over a Hermes installed outside Saka, the agent checks there is room for a copy of its data plus the image; if not, it says how much is needed and leaves the old Hermes running untouched. Earlier, a full disk was only found after the old Hermes had been stopped (then restored).
  • Interrupted tasksWhen the connection drops in the middle of a task, the message now says to reload the page and try again if nothing changed.
Oct 112026
featurefix4 updates#

Team members get alerts on their own Telegram

Teams can now share on-call: alerts reach each member's own Telegram for the servers they can access, with buttons that respect their role.

Team

  • Alerts for membersMembers who linked Telegram in their own account receive this account's server alerts, scoped to the servers shared with them. Combined multi-server alerts only go to members with access to all servers, so other server names are not exposed.
  • Buttons by roleViewers get Check again, View log, and Open in panel; Developers and Admins also get Restart. Membership, role, and server access are checked again when a button is pressed, member actions are recorded under their name, and the owner is told.
  • Members control their alertsA switch on the Team card turns a member's alerts on or off. The owner sees which members have linked Telegram. Team is now in the account menu and in the Start here steps.
  • FixViewer members could not switch back to their own account or leave a team from the panel; these now work for every role.
Oct 112026
feature3 updates#

Handle alerts from Telegram with buttons

Telegram alerts now carry buttons: view an app's log, restart it, check a site again, or open the panel, without leaving Telegram.

Telegram

  • Stopped appsView log sends the last 40 lines with tokens and passwords masked; Restart asks for confirmation, runs once, and is recorded in the action history.
  • Sites that do not openCheck again now checks from Singapore and, if needed, Bangalore and replies with the result.
  • Safe by designButtons carry a random token instead of server data, only work from the Telegram account linked to the owner, and expire after 7 days. Send /uji to the bot to see an example.
Oct 112026
feature2 updates#

Get alerts on Telegram with one click

Most alerts still went to email. The panel and every alert email now offer a one-click way to link Telegram.

Alerts

  • Telegram suggestionUntil Telegram is linked, a bar at the top of the panel shows how many important alerts went to your email and a Link Telegram button that opens the bot right away; the panel notices when you press Start. Press Later to hide it for 30 days (it returns if new alerts are emailed).
  • In alert emailsEvery alert email now ends with a Link Telegram button instead of a plain sentence.
Oct 112026
featurefix3 updates#

One message when many servers go down together, and help to connect a first server

Outages that hit several of your servers at once now arrive as a single message, and new sign-ups get one reminder with the three ways to connect a server.

Alerts

  • Combined outage messagesWhen 3 or more of your servers disconnect in the same check, or reconnect within a minute, you get one message listing them (usually a provider or network problem) instead of one per server. Telegram messages are paced to Telegram's limits and retried when Telegram asks to wait.

Getting started

  • Connect-a-server reminderAbout a day after signing up, verified users who have not connected a server get one email: run the one-line command on a VPS, create a server from a DigitalOcean, Vultr, or IDCloudHost account in the panel, or use a computer at home through a tunnel. It is sent once.

Monitoring

  • Faster checks at scaleInternet checks now run in parallel according to how many addresses are monitored (8 up to 96 at once), so a check round stays within a minute with thousands of sites.
Oct 112026
fixfeature2 updates#

Fewer false disconnect alerts, and Saka's own database restore-tested daily

Agent self-updates and control plane restarts no longer send server disconnected alerts, and the Saka control plane replica is restored and checked every day.

Alerts

  • Disconnected serversA server is reported as disconnected after 3 minutes without a heartbeat (was about 75 seconds), and not at all while its agent is updating itself or in the first minutes after the Saka control plane restarts. Short blips made up over a third of these alerts.

Reliability

  • Control plane restore testEvery night the standby control plane in Bangalore restores the live replica of Saka's database into a temporary folder and checks integrity, key tables, and how far it lags behind (seconds). The Saka team is alerted if it ever fails, so failover is proven to work before it is needed.
Oct 112026
fix2 updates#

Fewer false CPU and RAM alerts

Short CPU or RAM spikes that come and go are no longer reported as sustained load.

Alerts

  • Sustained onlyCPU and RAM alerts are only sent when usage stays high for the whole hold time (10 minutes for CPU, 5 for RAM). Before, three short spikes within 10 minutes were reported as if the server had been busy for 10 minutes, which on some servers meant dozens of alerts a day.
  • Restore test timingMonthly restore tests run between 01:00 and 06:00 in the server time zone; servers set to UTC now use WIB so tests do not run during Indonesian office hours. Tests are skipped when less than about 1 GB of RAM is free.
Oct 112026
feature2 updates#

Monthly backup restore tests, and uptime checked from two locations

Saka now proves each site's backup really restores, and only reports an outage when the site fails from both Singapore and Bangalore.

Backups

  • Restore testOnce a month Saka opens each site's latest backup in a temporary container with no internet access, imports the database, and loads the site. The Backups card shows restorable or failed with the reason, you are alerted on failure, and the monthly report counts sites proven restorable. Test restore now runs it on demand; the live site is never touched. MCP tool situs_uji_pulih.

Monitoring

  • Two locationsWhen a site, tunnel, load balancer, or external address fails from Singapore, it is checked again from Bangalore. Outages are only counted and reported when both fail, so a network blip at one point no longer causes a false alert.
Oct 112026
feature3 updates#

Choose how long S3 backups are kept, and backup health in reports

Set S3 backup retention straight from the server page so the bucket stops piling up, see how much is in the bucket, and get told when daily backups stop.

Backups

  • S3 retention you controlOn the Backup to S3 card press Change next to Kept for and pick 7 to 365 days or your own number (1 to 3650), without re-entering keys. Shortening asks first, then removes older archives right away; the newest backup of each site is always kept. MCP tool cadangan_s3_masa_simpan.
  • What is in the bucketThe card shows how many archives Saka keeps in the bucket and their total size.
  • Backup healthYou get an alert when a site's daily backup has not succeeded for more than 26 hours, and the monthly report shows sites backed up, the last S3 upload, and retention per server.
Oct 112026
feature4 updates#

Monitor any URL, monthly reports, and bandwidth per site

Watch addresses outside Saka every minute like an uptime service, get a monthly report by email and Telegram (and send each client their own site's report), and see each site's bandwidth on its page.

Monitoring

  • Addresses outside SakaSettings has a Monitor addresses outside Saka card: any URL or API, checked from the internet every minute, with an optional keyword or expected HTTP status. Outages over 3 minutes and certificates with less than 14 days left are reported; addresses can go on public status pages. Internal network addresses are refused. API /api/v1/pantau-luar, MCP tool pantau_luar.

Reports

  • Monthly reportOn the 1st at 08:00 WIB: outages and bandwidth per server, availability per site and address, WordPress security score, expiring certificates, and alert counts. Full by email, short over Telegram; preview and send now from Settings. MCP tool laporan_bulanan.
  • Reports for clientsAdd up to 3 client emails on a site page; each client only gets that site's report, without your servers or other sites.

Bandwidth

  • Per siteEach site page shows its bandwidth this month and last month with a daily chart, and a single site's spike is reported with its name.
Oct 112026
feature3 updates#

Bandwidth quota alerts, and a move offer when a new VM connects

Set your provider's monthly bandwidth quota and Saka alerts you at 80% and 100%, and reports unusual daily spikes even without a quota. A new, empty VM now suggests moving apps from your other servers to it.

Servers

  • Bandwidth quotaSet quota on the Bandwidth card (GB per month, in + out) shows a usage bar and sends one alert at 80% and one at 100% each month, over Telegram or email. API PATCH /api/v1/server/{id} with kuota_bw_gb, MCP tool server_kuota_bandwidth.
  • Spike alertsA day with at least 5 GB and five times this month's daily average is reported once, with the likely causes (bots scraping a site, large downloads or backups, a compromised server sending spam).

Moving to Saka

  • Offer on a new VMSaka no longer needs you to find the move button: when a new, empty VM connects and another of your servers runs Dokploy, Coolify, aaPanel, CyberPanel, OpenClaw or Hermes, the new VM's page shows a Move something to this server? card and you get one Telegram message. Detecting a platform alone never triggers an offer.
Oct 112026
feature3 updates#

Bandwidth per month and per site, and moving from CyberPanel

Every server now counts its bandwidth: in and out per day, per month, and per site, in the panel, the API, and for your AI agent. CyberPanel servers get the same one-click move as aaPanel.

Servers

  • Bandwidth usageA Bandwidth card on the server page shows data in and out for today, this month, and last month, with a 62-day daily chart, monthly totals, and usage per site (from the web access log). Counted every minute on the public network interface; internal Docker, VPN, and private network traffic is excluded and reboots do not lose the count. Figures start when the new agent starts counting. API GET /api/v1/server/{id}/bandwidth, MCP tool server_bandwidth.

Moving to Saka

  • Move from CyberPanelSaka recognises CyberPanel, keeps its panel, OpenLiteSpeed admin, FTP, email and DNS ports open when Saka's firewall is on, and moves its sites and child domains to another Saka server with their main database and domains while the site keeps running. MCP tool dokploy_pindah with platform cyberpanel.
  • Dokploy databases into a cluster, testedStandalone Dokploy databases were copied into a Saka cluster through the temporary path on Dokploy's network. A database newer than the cluster (for example PostgreSQL 18 into 17) is refused with a clear message.
Oct 102026
feature4 updates#

Move from aaPanel, standalone databases into a cluster, and Compose apps that move and deploy on push

aaPanel servers get the same one-click move as Dokploy and Coolify, databases no app uses can be copied into a Saka database cluster, and Docker Compose apps now deploy on every push and can move to another server.

Moving to Saka

  • Move from aaPanelSaka recognises aaPanel, keeps its panel, phpMyAdmin and FTP ports open when Saka's firewall is on, and moves its PHP, WordPress and static sites to another Saka server: files, the main database and domains, copied server to server while the site keeps running. PHP below 8.2 runs on 8.2 with a note. MCP tool dokploy_pindah with platform aapanel.
  • Standalone databases into a clusterPostgreSQL, MySQL and MariaDB databases on Dokploy, Coolify or aaPanel that no Git app uses can be copied into a Saka database cluster from the same move dialog. Saka opens a temporary path on 127.0.0.1 of the old server only, copies in one transaction, compares row counts, and closes the path.

Docker Compose

  • Deploy on pushCompose apps from Git get a webhook (GitHub, GitLab, Gitea): every push to the branch pulls the latest code and restarts the services.
  • Move to another serverCompose apps use the regular Move to another server card: files, variables, the deploy key and volume data are copied to the new server and the local port is adjusted. Services pause briefly while their volumes are packed.
Oct 102026
feature4 updates#

Move over from Coolify

Servers running Coolify get the same one-click move as Dokploy: apps, services, compose volumes and the databases your apps use land on another Saka server, and Coolify keeps running untouched.

Coolify

  • Move from CoolifySaka reads Coolify's apps straight from its database on the server (no API key; variables are decrypted locally with Coolify's own APP_KEY) and recreates them on another Saka server: repo apps (Nixpacks, Dockerfile, static) as Git apps, Docker image, Docker Compose and one-click services as Compose apps with their volume data, and standalone databases used by a Git app with the same host name, credentials and data. Shared variables are filled in. MCP tool dokploy_pindah with platform coolify.
  • New domain, updated variablesWhen you pick a new domain while moving (Coolify or Dokploy), variables that mention the old domain, such as SERVICE_URL_* or APP_URL, point to the new one.

Sites

  • Compose volumesNamed volumes used by a service but not declared at the top of the compose file are declared automatically, so templates written for Coolify run as is.
  • Catalog logosPocketBase, Cal.com, Directus and Supabase show their official logos.
Oct 102026
feature5 updates#

Databases for Git apps, Dokploy databases move too, and Supabase, Cal.com, PocketBase, Directus

Git apps on a server get a one-click database, Dokploy apps now bring the databases they use when moved to Saka, and the catalog adds Supabase, Cal.com, PocketBase and Directus.

Sites

  • App databaseAdd PostgreSQL, MySQL, MariaDB, MongoDB or Redis to a Git app in one click. It runs on the same server on the app's private network (no port open to the internet), DATABASE_URL and friends are set automatically, and its data is in daily backups and restores. MCP tool situs_git_database.
  • Catalog: SupabaseThe full Supabase stack (PostgreSQL, auth, APIs, storage, realtime, Studio) from its official repo. Saka downloads only the docker folder, creates every secret on the server, signs the anon and service_role keys, and shows the Studio login. Needs 4 GB RAM.
  • Catalog: Cal.com, PocketBase, DirectusAppointment scheduling with its own PostgreSQL, a one-file backend and a no-code CMS. PocketBase and Directus get an admin account with a random password, shown with Show login.
  • Compose namescontainer_name in a compose file becomes a network alias, so services that call each other by container name keep working.

Dokploy

  • Databases move tooA Dokploy database used by a Git app moves with it: same image, credentials, host name and data path, so the old DATABASE_URL works unchanged. Its data is copied server to server; the Dokploy database pauses briefly.
Oct 102026
featuresecurity6 updates#

Docker Compose apps, and moving over from Dokploy

Run any docker-compose.yml as a Saka app with HTTPS and daily backups including volumes. Servers running Dokploy or Coolify are recognised and left alone, and Dokploy apps can be moved to another Saka server with their variables, domains and compose volumes.

Sites

  • Docker ComposePaste a docker-compose.yml or use one from a Git repo, pick the web service and its port, and Saka serves it on your domain with HTTPS. The Compose plugin is installed automatically.
  • Safe by defaultSaka rewrites the file before running it: other published ports bind to 127.0.0.1 so databases are not exposed, container_name is dropped, restart defaults to unless-stopped. host networking, privileged, the Docker socket and system folders are refused with an explanation.
  • Backups with volumesDaily and manual backups include every named volume (services pause briefly), and restore puts the volumes back. Variables fill ${NAME} and are written to .env for env_file.

Dokploy & Coolify

  • Left aloneSaka detects Dokploy and Coolify, keeps their ports and Docker Swarm ports open when you turn on Saka's firewall, and refuses Saka sites and Kubernetes on that server with a clear reason. OpenClaw, Hermes, databases, VPN and Shield still work there.
  • Move from DokploySaka reads Dokploy's apps straight from its database on the server (no API key; encrypted variables are decrypted locally) and recreates them on another Saka server: Git apps as Saka Git apps, image and compose apps as Compose apps, with variables, domains and compose volume data copied server to server. Dokploy is not changed. Standalone Dokploy databases are listed but not moved yet. MCP tool dokploy_pindah.

VPN

  • Fewer accidentsThe device table is tidier, and Remove the VPN from this server moved under Advanced settings and asks you to type the server name; revoking one device stays one click on its row.
Oct 102026
feature3 updates#

Move OpenClaw or Hermes to a cheaper VPS

Found a cheaper or faster VPS? Install Saka on it and move OpenClaw or Hermes there in one click, with its memory, sessions, settings, AI keys and channels. It also works for assistants you installed yourself before Saka.

Apps

  • Move to another serverWorks for Saka-managed apps and self-installed ones (Docker or the official installer), which Saka detects automatically. Data goes straight between the two servers over a one-time certificate, limited to the new server's IP; Saka Panel never sees it.
  • Short pause, never two botsThe new server is prepared first while the old bot keeps running. The old bot then pauses (about 30 seconds for 18 MB in our test), its data is sent and checked, and the bot starts on the new server. Two bots with one Telegram token never run at the same time.
  • Safe on failureIf any step fails, the old bot starts again automatically. Brief network drops are ridden out, and a move interrupted by a Saka restart is finished or rolled back. On success the old bot is stopped, not deleted. MCP tool aplikasi_pindah.
Oct 102026
featurefix7 updates#

One Deploy app button, code upload, databases and branch previews on Kubernetes

Deploying to Kubernetes now starts from one button: upload your code, connect Git, use a Docker image or a Helm chart. Apps built by Saka can get their own PostgreSQL, MySQL or Redis in one click, and every branch can be tried at its own address.

Kubernetes

  • One Deploy app buttonChoose where the app comes from: Upload code, From Git, Docker image or Helm chart. Apply YAML moved under Advanced. Suggested by a Kubernetes user.
  • Upload codeUpload a .zip, .tar.gz or .tar (up to 300 MB). It streams straight to the cluster's control plane and is built there; Saka Panel never stores your code. A single top-level folder is unpacked automatically. Upload new code any time; databases, variables, scaling and rollback work as with Git.
  • One-click databasePostgreSQL, MySQL (MariaDB) or Redis runs in the cluster just for that app. Saka creates the password inside the cluster (it never leaves it) and fills in DATABASE_URL, DB_HOST, DB_DATABASE, DB_USERNAME, DB_PASSWORD or REDIS_URL automatically. One SQL database plus Redis per app.
  • No accidental data lossAn app cannot be deleted while it still has a database, and deleting a database asks you to type its name.
  • Branch previewsTurn on previews and every push to another branch builds its own copy at https://<branch>-<app>.<ip>.sslip.io. Previews never connect to the app's database. They disappear when the branch is deleted or after 7 days without a push; up to 5 per app.

Sites

  • Safer migrationsWordPress sites moved into Saka now block PHP in the uploads folder by default, like new installs, and get their first security scan within 30 minutes.

Apps

  • Hermes updatesHermes switched from date versions to v0.21.x; the update check failed every night. Saka now understands both.
Oct 102026
featuresecurity5 updates#

Find online gambling injections and backdoors on your sites

Saka now checks WordPress and PHP sites every day for the hacks that hit Indonesian sites most: hidden slot/togel pages shown only to Google, gambling spam in posts, PHP backdoors and web shells. One click moves suspicious files to a quarantine you can undo.

Sites

  • Gambling injection scanSite files, the home page as Google sees it versus a normal visitor (cloaking), WordPress posts and settings, and admins created in the last 14 days. Findings appear in the security card and on Telegram.
  • BackdoorsPHP files with obfuscated eval, input-driven function calls and known web shells, plus any PHP file inside wp-content/uploads.
  • QuarantineMoves the flagged files out of the web folder. If the site breaks afterwards, every file is put back automatically. WordPress core and wp-config.php are never moved (use Reinstall original files). Quarantined files can be restored.
  • PHP sites tooPlain PHP sites get the same daily check and quarantine, not just WordPress.

Servers

  • Crypto minersServer security flags crypto miners and programs running from /tmp, /dev/shm or /var/tmp, and one button stops them, deletes the file and removes the cron jobs that restart them.
Oct 102026
featurefix7 updates#

Deploy apps to Kubernetes straight from Git

Point Saka at a GitHub, GitLab or Gitea repository and it builds the image inside your cluster and runs it with HTTPS. No registry account, no Dockerfile needed, no YAML.

Kubernetes

  • Apps from GitBuilds run in the cluster with BuildKit, using your Dockerfile or detecting the stack automatically (Node, Python, PHP, Go and more). Images go to a private registry inside the cluster.
  • Private reposSaka creates a deploy key for private repositories; add it to the repo once.
  • Push to deployA webhook redeploys on every push to the chosen branch, signed per app.
  • Releases and rollbackThe last 10 releases are kept; roll back with one click. Variables are stored as a Kubernetes Secret, not in the image.
  • Copies and autoscaleChoose how many copies run, or let Saka add copies when CPU is busy and remove them when it is quiet, in plain words instead of HPA settings.
  • Pod limit by RAMNew clusters allow 160 or 250 pods on servers with 8 or 16 GB RAM instead of the fixed 110, and existing clusters get a Raise pod limit suggestion that applies it one node at a time.

Servers

  • Package lockInstalling Docker or other packages while the system is updating no longer fails; Saka waits for the other install to finish.
Oct 92026
fix4 updates#

Agents reconnect in seconds after network glitches

After a brief network outage, some agents kept a dead connection open and showed as disconnected for 15 minutes or more. They now notice within about 30 seconds and reconnect on their own, and a cluster upgrade waits through short outages instead of stopping.

Servers

  • Dead connection detectionThe agent pings Saka every 30 seconds and reconnects when there is no answer within 20 seconds, instead of waiting for TCP to give up.

Kubernetes

  • Upgrades ride out glitchesWhen a node's agent drops for a moment during an upgrade, Saka waits up to 3 minutes for it to come back and retries, instead of stopping the upgrade.
  • Page keeps loadingThe cluster page shows the last known status when the control plane answers slowly, instead of staying on Loading.
  • Safer releasesSaka releases are refused while a cluster upgrade, node change or VPN setup is running.
Oct 92026
feature5 updates#

Kubernetes for people who never used Kubernetes

See how much room the cluster has left, add servers without technical steps, and get plain-language fixes for the traps that confuse beginners: the hidden pod limit, stuck nodes and piles of failed pods.

Kubernetes

  • Capacity cardOverview shows total, used and free CPU, RAM, disk and pod slots, what apps have reserved, and the space left for volumes, in plain words. Node rows show disk and pods per server.
  • Add capacityOne button creates new servers in your cloud account on the cluster's private network and joins them as workers. Bare metal and other servers still join through Add node.
  • Pod slotsKubernetes limits pods per server (110 by default) separately from CPU and RAM. Saka now counts slots, warns before they run out, and no longer mistakes Too many pods for missing CPU/RAM.
  • Stuck nodesHealthy nodes left cordoned get an Allow again button; nodes whose pod network is not ready are explained instead of silently ignored.
  • Failed podsFailed and evicted pods older than 24 hours are cleaned up automatically every 6 hours, with a Clean up now button. Running apps are not touched. MCP tools k8s_rawat and k8s_tambah_kapasitas.
Oct 92026
fix3 updates#

Safer clicks around the VPN

Removing a server now cleans up its VPN and never leaves sites locked behind a VPN that no longer exists, and closing a whole site to the public needs the domain typed in.

VPN

  • Release cleans upRelease server now removes the team VPN (WireGuard, gateway DNS, firewall port) and lists it in the release plan.
  • No locked sitesWhen the server hosting a VPN is removed, sites on any server that were VPN only through it are opened again first.
  • Type to confirmMaking a whole site VPN only explains that customers, search engines, webhooks and payments will get 403, and asks you to type the domain.
Oct 92026
featurefix3 updates#

Sites that open only through your team VPN

Restrict WordPress admin, specific paths, or a whole internal app so only devices on your Saka VPN can open it, with split-mode devices routed automatically.

VPN

  • VPN onlyNew VPN only card on every site: protect the WordPress admin (admin-ajax.php stays open), specific paths such as /admin*, or the whole site. Visitors outside the VPN get a 403 page; it works behind Cloudflare and load balancers. Removing a VPN lifts its protection, so nothing is ever locked out. MCP tool situs_hanya_vpn.
  • Split mode just worksSplit devices created after protection is on include protected sites automatically. Sites on the VPN server itself are reached through a small DNS on the VPN gateway, because routing the server's public IP into the tunnel would break it.

Servers

  • Firewall keeps sites upTurning on the server firewall now always keeps ports 80 and 443 open when the server runs Saka sites, and the VPN port when it is a VPN gateway.
Oct 92026
feature7 updates#

Team VPN, activity log, safer app updates, and an app check for AI-built apps

Turn a server into a WireGuard VPN for your team, see who changed what (AI agents included), update catalog apps knowing what changed and roll back within 7 days, catch secrets and open APIs in Git apps, and free disk space automatically when it runs low.

VPN

  • Team VPNNew VPN menu: Saka installs WireGuard on a server with a public IP. Each laptop or phone gets its own key as a QR code or .conf file for the free official WireGuard app, with step-by-step guides for Windows, Mac, iPhone, Android and Linux.
  • Split or fullSplit mode sends only internal systems (the server's private network plus extra IPs, CIDRs or domains) through the VPN; full mode sends everything, so the device goes out with the server's IP for bank or partner IP whitelists.
  • ControlDevices can expire, are revoked instantly, show last connection, source IP and data, and you get a Telegram message when a device connects for the first time. Device secret keys are never stored by Saka. MCP tool vpn.

Account

  • Activity logAccount menu, Activity log: every change with who did it (you, a team member, an API token by name, or an AI agent through MCP with the tool name), target, source IP and result. Filter by AI agents, team, API or panel; kept 180 days. MCP tool riwayat_aksi.

Sites

  • Catalog app updatesSaka checks once a day whether a newer image is available (without pulling it), shows the version or image date with release notes, and tells you once per new version. After an update the previous version is kept for 7 days with a Roll back button.
  • App checkEvery Git deploy checks for secrets committed to the repo (.env files, cloud, payment and AI API keys, database URLs with passwords, Supabase service_role), env variables the code reads but are not set, and /.env, /.git or user APIs open without login. Never blocks a deploy; important findings go to Telegram once per commit.

Servers

  • Automatic disk cleanupWhen a disk passes 90%, Saka immediately removes what is safe to remove (unused Docker images, build and package caches, old system logs) and tells you how much was freed.
Oct 92026
featurefix4 updates#

Kubernetes cluster page in tabs, language saved to your account, and buttons that failed silently

The cluster page is split into Overview, Apps, Nodes, Data & backups and Access, four panel buttons that were rejected by the CSRF check work again, and the panel no longer sticks to an old version after a deploy.

Kubernetes

  • TabsThe cluster page has five tabs. Overview shows nodes Ready, apps not ready, domains and the latest snapshot, each linking to its tab. The active tab is kept in the URL, so links and reloads open the same tab. Progress and errors stay above the tabs.

Panel

  • Buttons fixedTurn on fixed API address, revoke kubectl access, check Git repo access and delete preview were sent without a JSON body and rejected with 'must be application/json'. They now work.
  • Fresh after deployPanel pages are served with Cache-Control: no-cache, so browsers check for a new version instead of reusing an old page.
  • Language follows your accountChoosing ID or EN is saved to your account, so the panel opens in the same language on every device and browser. Team members keep their own choice.
Oct 92026
featurefix6 updates#

Safer node reboots, a free test address for new sites, and a calmer server page

Rebooting a Kubernetes node now moves its pods away first, new sites can start on a free test address without a domain, and the server page drops noisy or contradictory messages.

Kubernetes

  • Safe node rebootReboot and the pending-reboot Fix button on a Kubernetes node now cordon and drain the node first, reboot it, wait until it is Ready and let it accept pods again. If pods cannot be moved, the reboot is cancelled and the node keeps running.
  • Clearer node statusInformational k3s conditions such as EtcdIsVoter no longer show as problems; firewall status on nodes consistently says it is managed by the cluster.

Sites

  • No domain yet?The create-site form offers a free test address (name.IP.sslip.io) that works immediately with HTTPS and no DNS setup; switch to your own domain later.

Servers

  • Task historyAutomatic agent updates are hidden from the task history by default (one click shows them), and tasks show readable names instead of command names.
  • ReadinessNo more 'OpenClaw needs 8 GB' when OpenClaw is already installed, and no app advice on Kubernetes nodes.
  • Smaller fixesKubernetes nodes are no longer offered for tunnels, the browser tab title follows the panel language and page, and the add-server requirements list WordPress and Kubernetes too.
Oct 92026
fix2 updates#

More reliable agent updates and clock sync

Agent updates now download from a storage bucket instead of the control plane, with retries and a fallback, and servers whose provider blocks NTP get their clock set from HTTPS time.

Servers

  • Agent updatesUpdates come from a public storage bucket (still verified by the Ed25519 release signature), with the control plane as a fallback and up to two tries per source. Servers that cannot reach the bucket are switched to the control plane automatically. A full disk is reported clearly and retried later instead of failing silently.
  • Clock syncWhen the provider blocks NTP (UDP 123), Fix clock sets the time from HTTPS instead and Saka keeps correcting it every 6 hours until NTP works again.
Oct 92026
featurefix3 updates#

Kubernetes app domains keep working when a node dies

Every cluster gets an ingress address that always points at healthy nodes, the panel checks each app domain's DNS, and Traefik retries requests away from dead nodes in 3 seconds.

Kubernetes

  • Ingress addressapp-<id>.k8s.saka.work has an A record for every node that is Ready, connected and answering on port 80, checked every minute. Point subdomains at it with a CNAME; root domains get A records for all IPs. Replaces the old advice of pointing a domain at any single node. New MCP tool k8s_domain.
  • Domain checkEach Ingress domain shows whether it is a CNAME to the cluster, points at all nodes, only some nodes, somewhere else, or nowhere yet, using a public resolver.
  • Faster retriesTraefik now gives up on a silent backend after 3 seconds instead of 30, so the retry middleware moves to a live pod quickly. Applied automatically to existing clusters.
Oct 92026
feature2 updates#

kubectl keeps working when a control plane dies

Kubeconfigs now point at a fixed address per cluster with an A record for every healthy control plane, so kubectl, CI and AI agents keep working when one control plane is down.

Kubernetes

  • Fixed API addressEach cluster gets <id>.k8s.saka.work, included in every control plane's certificate. Saka keeps its A records in sync with the control planes that are Ready, every minute. New clusters use it automatically; older clusters can turn it on with a rolling restart of the control planes.

Panel

  • Enterprise in the account menuThe Enterprise form moved from the top bar into the account menu.
Oct 92026
feature2 updates#

Kubernetes volumes that survive a node failure (Longhorn)

Turn on replicated volumes for a Kubernetes cluster: every volume is copied to up to 3 nodes, and when a node dies its pods move to another node together with their data.

Kubernetes

  • Volume storageOne click installs open-iscsi on every node and Longhorn 1.12, optionally as the default storage class. Pods on a dead node are moved and their volumes reattached automatically; missing copies are rebuilt when the node returns. New nodes are prepared automatically. New MCP tool k8s_penyimpanan.
  • Volume healthEach volume shows healthy, missing copies or failed, with Telegram alerts when a volume is missing copies for 30 minutes or fails.
Oct 92026
feature2 updates#

Private repos ask for the deploy key first; deploy apps to labeled nodes

New apps from a private Git repo no longer fail their first deploy: Saka checks access first and walks you through adding the deploy key. Kubernetes apps can be pinned to labeled nodes from the deploy dialog.

Sites

  • Deploy key firstFor git@ repos Saka checks access before the first deploy. Until the key is added nothing is deployed; you get the key by Telegram or email, and the app page shows a Last step card with Check & deploy and a link to the guide for GitHub, GitLab, Bitbucket and Gitea.

Kubernetes

  • Run on labeled nodesThe Deploy app dialog lists your node labels; picking one sets a nodeSelector and tolerates the taints on those nodes automatically. Also available as node_label and toleransi in the k8s_aplikasi MCP tool.
Oct 92026
featuresecurity3 updates#

Rolling k3s upgrades with automatic rollback

Upgrade a Kubernetes cluster to the latest k3s patch or the next minor from the panel: snapshot first, one node at a time, verified binaries, and automatic rollback if a node fails.

Kubernetes

  • Upgrade availableThe cluster page offers the latest patch of the current minor and the next minor, based on the oldest node so a stopped upgrade can be resumed. New MCP tool k8s_upgrade.
  • Safe by defaultetcd snapshot first; control planes before workers; each node is drained, gets the official k3s release verified by SHA256, and must be Ready on the new version before the next one. A failing node is rolled back to the binary it had before this upgrade and the upgrade stops.
  • Pick a versionNew clusters can be created on a specific k3s version, for example to match another environment.
Oct 92026
featurefix3 updates#

Rebuild a Kubernetes cluster from S3 when all nodes are gone

Lost every server of a cluster? Rebuild it on new servers with only the S3 recovery code, either with apps and volume data from Velero or with the exact cluster state from an etcd snapshot.

Kubernetes

  • Restore from backupPick new empty servers and paste the recovery code; the new server lists the clusters in your bucket. Velero method: a new cluster plus all app namespaces and volume data, read through a read-only location so old backups are never overwritten. etcd method: every resource including the CA, so old kubeconfigs keep working; old nodes are removed. New MCP tool k8s_cari_cadangan.
  • Token backed upThe cluster token (needed to restore etcd on new nodes) and k3s version are now uploaded encrypted next to the snapshots.
  • 10.43.x networksClusters on servers whose private IP is inside k3s's default pod/Service ranges (such as IDCloudHost's default 10.43.x network) now use 10.52/10.53 so Service IPs cannot collide with server IPs.
Oct 92026
feature2 updates#

Kubernetes alerts on Telegram

Saka now tells you when a Kubernetes node goes NotReady or runs out of resources, pods are stuck or crash-looping, HTTPS certificates are not issued, or etcd snapshots stop, and again when it recovers.

Kubernetes

  • What is watchedNode NotReady (90 s), node memory/disk/PID pressure (2 min), pods waiting for CPU/RAM (5 min), crash-looping, OOMKilled or image-pull failures (5 min), certificates not issued after 15 minutes, and no etcd snapshot for a day.
  • Quiet when expectedAlerts are held while nodes are being added or removed or the cluster is being restored, and a server that is disconnected from Saka only gets the server alert.
Oct 92026
feature5 updates#

Kubernetes: scoped kubectl access, Helm charts, node labels & taints, layout advice

Revocable kubeconfigs with limited roles for people and AI agents, Helm charts without the helm binary, labels and taints per node, advice on when to add control planes or workers, and a node summary on the server page.

Kubernetes

  • kubectl accessCreate kubeconfigs per person, CI or AI agent with a Viewer, Developer (optionally one namespace) or Admin role and an expiry of 1 to 365 days. Each is its own ServiceAccount with a bound token; revoking it rejects the token immediately. Saka never stores the token. New MCP tool k8s_akses.
  • Helm chartsInstall and upgrade charts from https repos or OCI registries with your own values through k3s's built-in Helm controller; failures show the Helm log. New MCP tool k8s_helm.
  • Labels & taintsSet node labels and taints when adding a node (taints apply from registration) or later from the node row. Kubernetes and k3s keys are protected.
  • Layout adviceThe cluster page explains when to add a control plane (1 or an even number of them) and when to add a worker (pods waiting for CPU/RAM, high reserved or real usage, busy control planes), with a button that picks the right role.
  • Node summaryServer pages of Kubernetes nodes show node status, CPU/RAM usage, labels, taints and the pods on that node.
Oct 92026
feature2 updates#

Restore a Kubernetes cluster from an etcd snapshot

A Restore cluster button returns the whole cluster state to a local or S3 etcd snapshot, with an automatic undo snapshot first.

Kubernetes

  • Restore clusterPick a local or S3 snapshot from any control plane and type the cluster name. Saka saves a sebelum-pulih snapshot, stops all control planes, resets etcd from the snapshot (downloading and decrypting it from S3 when needed), then rejoins the other control planes. Old etcd data is set aside, not deleted.
  • Safe by defaultAll control planes must be connected; if a step fails, the cluster is started again on its old data. Node changes and restores cannot overlap. New MCP tool k8s_pulihkan.
Oct 92026
feature3 updates#

Add and remove Kubernetes nodes, VM or bare metal

Grow a Kubernetes cluster with new workers or control planes from any connected server, VM or bare metal, and remove nodes safely with drain.

Kubernetes

  • Add nodePick an empty connected server and a role on the cluster page. Saka joins it with the cluster's own k3s version and network mode; a single-private-network cluster refuses servers outside that network with an explanation instead of a broken node. Bare metal works the same as a VM once the Saka agent is installed.
  • Remove nodePods are drained to other nodes, the node is deleted from the cluster, then k3s is uninstalled; the server stays in Saka as a regular server. The last control plane cannot be removed.
  • One network mode per clusterNew clusters record their network mode (vxlan on one private network, WireGuard across networks) and every node uses it, so mixed-network nodes no longer get a mismatched backend. New MCP tool k8s_node.
Oct 92026
feature3 updates#

Kubernetes app volume backups with Velero

Opt-in backups of app volumes (databases inside pods, uploads) to the cluster's S3, encrypted with a password derived from the recovery code, with one-click restore.

Kubernetes

  • Velero, installed for youTurn on volume backups on the cluster page: Saka installs Velero 1.18 with the S3 plugin and node agents through the k3s HelmChart, pointed at the cluster's S3 destination. Daily at 02:00 and Back up now; retention follows the S3 setting.
  • Encrypted with your recovery codeVelero's file-system backups are encrypted by a repository password; Saka derives it from the key in the recovery code instead of Velero's shared default, so the code alone can open the backups even if the whole cluster is gone.
  • RestoreRestore a backup from the list; existing resources are not overwritten. New MCP tool k8s_volume.
Oct 92026
feature3 updates#

Kubernetes backups are per cluster, encrypted to your own S3

etcd snapshots of a Kubernetes cluster can now be sent encrypted to your own S3 from every control plane, set up once on the cluster page.

Kubernetes

  • Per cluster, not per VMOne S3 setting on the cluster page is installed on every control plane with the same encryption key. Each control plane uploads its etcd snapshots (automatic every 12 hours plus Back up now) encrypted with age, so backups survive even if all nodes are lost. Retention follows the S3 setting, always keeping the newest four.
  • Recovery codeThe same recovery code as site backups; keys stay on the nodes and are never stored by Saka.
  • Node pagesThe per-VM site backup card is hidden on Kubernetes nodes.
Oct 92026
feature2 updates#

Saka finds k3s clusters you have not adopted yet

The Kubernetes page now offers to adopt k3s clusters already running on your servers, grouped as one cluster instead of separate control planes.

Kubernetes

  • Found clustersAgents report a fingerprint of the cluster CA, so nodes of the same cluster are grouped. An empty Kubernetes page shows the cluster with its control planes and worker count and an Adopt button. New MCP tool k8s_temuan.
  • Simpler adoptionThe adopt form lists whole clusters (k3s cluster, 6 nodes) instead of each control plane.
Oct 82026
feature8 updates#

Kubernetes (k3s) on your own servers, managed from Saka (beta)

Create a k3s cluster from empty servers or adopt one you already run, deploy container apps with automatic HTTPS without writing YAML, and manage it all from the panel or your AI agent.

Clusters

  • Create or adoptPick empty servers: 1-2 give one control plane, 3 or more give three (keeps running when one dies) plus workers. Nodes join over the private network when there is one, otherwise over WireGuard. Existing k3s clusters can be adopted from one control plane.
  • Tested defaultsTraefik on every node instead of a single copy, cert-manager with Let's Encrypt, dead nodes detected in 20 seconds and pods moved 30 seconds later, and retry on the ingress. In tests on IDCloudHost with six nodes, losing a control plane cost 1% of requests over 14 seconds.
  • No control plane feeOnly the VMs are paid for. The join token passes straight from the first node to the others and is not stored by Saka; neither is a kubeconfig.

Apps

  • Deploy appImage, port, replicas, RAM limit, domain, and variables. Saka builds the Deployment (replicas spread across nodes), Service, and Ingress with HTTPS. Deploy again with the same name to update.
  • YAML, logs, deleteApply any manifest, read workload or pod logs, see failing pods with their reason, and delete resources. System components are protected.
  • etcd snapshotsAutomatic every 12 hours plus a Back up now button.

For AI agents

  • MCP toolsk8s_daftar, k8s_status, k8s_buat, k8s_adopsi, k8s_aplikasi, k8s_terapkan, k8s_hapus, k8s_log, and k8s_cadangan; changes need the user's approval.
  • FixedThe wp_kuatkan, berkas_*, and email_dns_cek tools answered 404 because their paths were missing from the MCP router; a new test now checks every tool path.
Oct 82026
fix2 updates#

Kubernetes nodes are no longer treated as empty VMs

Servers that are already Kubernetes nodes (k3s or kubelet) are recognized, and VM-style actions that could break the cluster are hidden and refused.

Servers

  • Kubernetes node cardThe server page shows the node role (control plane or worker) and explains that sites and apps run as cluster workloads. The server list shows a Kubernetes badge.
  • Protected clusterCreate site, install app, firewall, swap, and moving web ports are hidden in the panel and refused by the agent on Kubernetes nodes, since they would fight the cluster's ingress for ports 80/443 or close ports between nodes. Monitoring, Server security, Shield, and system updates keep working.
Oct 82026
fix2 updates#

Fixes for servers behind NAT (IDCloudHost) and cloud API rate limits

Sites on IDCloudHost servers no longer show DNS as not pointing when it is, and creating several servers in a row no longer fails on the provider's rate limit.

Servers behind NAT

  • Public IP detectedIDCloudHost maps the public IP to the VM through NAT, so the VM itself only has a private address. The agent now asks Saka Panel which public IP it connects from and uses it for the site DNS check, the server IP on site pages, and Saka Shield's trusted list.

Cloud accounts

  • Rate limitsWhen a provider answers with a rate limit (IDCloudHost: API rate limit exceeded, or HTTP 429), Saka waits and retries up to three times instead of failing.
Oct 82026
feature2 updates#

Who Saka is for: four stages, from first site to Kubernetes

The home page now sorts Saka's services by the kind of user, so everyone can see where to start.

Home page

  • Four stagesSmall businesses and site owners, app builders, businesses that cannot stop, and tech teams on Kubernetes. Each stage lists the features to use with links to the guides, and the paid services it replaces.
  • Kubernetes is nextManaged k3s on your own VMs is marked coming soon. A proof of concept on three VMs passed: the cluster heals itself when a node dies, and with tuned defaults only 1.3% of requests failed during an 18-second window.
Oct 82026
feature3 updates#

Email DNS advice, and WordPress findings in English

The Outgoing email card now checks the sender domain's SPF, DKIM, DMARC, and MX and shows the exact records to add; WordPress security findings are now translated in the English panel.

Outgoing email

  • Check DNSReads SPF, DKIM, DMARC, and MX through public DNS (1.1.1.1) and suggests records matched to the email service in use: Brevo, Amazon SES, Google, Mailgun, SendGrid, Postmark, Mailjet, Zoho, Microsoft 365, Resend. SPF include and redirect chains are followed like receivers do; duplicate SPF records and +all are flagged.
  • Advice onlyDNS stays with the user's registrar or Cloudflare; Saka names where it is managed and gives copy-ready values but changes nothing. New read-only MCP tool email_dns_cek.

English panel

  • WordPress findingsVulnerability, update, closed plugin, and file integrity findings, their advice, and the source line now appear in English, without touching user content such as pull request titles.
Oct 82026
featuresecurity9 updates#

Security score for WordPress and servers, plus a file manager

WordPress sites get a 0-100 security score with known CVEs and one-click hardening, every server gets a Server security card, and sites get a file manager with a 7-day trash.

WordPress security

  • Known vulnerabilities (CVE)Installed WordPress, plugin, and theme versions are matched daily against WPVulnerability.net (free, no API key). Each finding lists CVE numbers, severity, and the fixed version; vulnerabilities with no fix yet are flagged so an active plugin can be deactivated.
  • Hardening with Fix buttonsPlugin/theme file editor, debug mode, unique salts, wp-config.php permissions, XML-RPC, PHP in uploads, and the default role for sign-ups, each fixable with one click (replacing salts and closing XML-RPC ask first). An admin called admin is flagged with the steps to replace it.
  • ScoreFindings and hardening add up to a 0-100 score shown as a ring on the WordPress card. New MCP tool wp_kuatkan.
  • Safer new installsNew WordPress sites from Saka get a random admin username instead of admin, wp-config.php readable only by the site, and PHP blocked in the uploads folder. The file editor and XML-RPC stay your choice.

Server security

  • Checklist and scoreA new card on the server page checks SSH password login, root password login, firewall, pending security updates, automatic security updates, pending reboot, and Saka Shield.
  • Safe fixesSSH keys only is applied only when an SSH key is already installed, validated with sshd -t, and rolled back if the reload fails; connected sessions stay open. Automatic security updates use unattended-upgrades or dnf-automatic with no automatic reboot. Firewall, updates, and reboot reuse the existing flows.

File manager

  • cPanel File Manager replacementBrowse, edit text files (Ctrl+S), create, rename, upload up to 100 MB, extract zip, and download files or whole folders as zip, streamed straight from the server. Git apps open their /data folder.
  • Nothing lost by accidentDelete moves to a 7-day trash; every edit or overwrite keeps the previous version there, restorable with one click. All access goes through a root-confined handle, so paths, symlinks, and zip entries cannot reach outside the site folder.
  • For AI agentsberkas_daftar, berkas_baca, and berkas_ubah (with approval).
Oct 82026
feature6 updates#

Git apps keep their data, get backups, and can send email

Apps from Git now have a persistent /data folder that survives deploys and is backed up daily, secrets can be generated on the server, and an Outgoing email card with a real test send.

Apps from Git

  • Persistent data folderEvery Git app gets /data (the DATA_DIR variable) that is kept across deploys, rollbacks, and variable changes, shared with workers and cron. SQLite databases and uploads made by AI-built apps no longer vanish on the next push. Any files the image already shipped in /data are copied in the first time, and the folder is owned by the app's user.
  • Backups for Git appsOnce the data folder has content it is backed up every day (last 7 kept), with manual backup, download, restore, and your own S3 like any other site. The app is frozen for a few seconds during the backup so SQLite is copied intact; restore restarts the app on the restored data.
  • Previews and two locationsPull request previews get their own copy of the data folder (under 1 GB). A second location is refused while the data folder holds SQLite, with a pointer to the MariaDB cluster.
  • Random secretsType a variable name and press Random: a 64-character secret is generated on the server and never shown. AI agents use situs_env acak, so SESSION_SECRET and the like never pass through chat.
  • Outgoing emailA new card stores your email service's SMTP settings as SMTP_* variables and sends a test email from the app's own server. Clear errors for blocked ports on new VPS accounts, wrong passwords (with the Gmail App Password hint), unverified senders, and sandbox recipients. New MCP tool situs_email_uji.
  • For AI agentspanduan_deploy now tells agents to keep SQLite and uploads in DATA_DIR, read SMTP_* for email, and generate secrets with acak.
Oct 62026
feature5 updates#

Servers that already run CasaOS, nginx, or Apache now work with Saka

A VPS whose port 80 is taken by CasaOS, nginx, or Apache no longer hits a dead end: one Set up button moves the old service aside and puts Saka in front of it.

Servers

  • One button, nothing deletedThe Readiness card has a new Web ports (80/443) row. Set up shows a plain-language plan, backs up the old service's settings, moves CasaOS, nginx, or Apache to another port (e.g. 8081), and starts Saka on ports 80/443. Every address that is not a Saka site is forwarded to the old service, so the CasaOS dashboard or the old site stays reachable from the server address as before.
  • Safe by designAny failed step restores the original settings automatically; releasing the server puts the old service back on port 80. If the old service also uses port 443, or a Docker container or another program holds the port, Saka changes nothing and explains why.
  • Clearer errorCreating a site on such a server now points to the exact fix instead of only saying the port is in use.
  • For AI agentsserver_perbaiki kinds web-cek (plan) and web (needs approval), mentioned in panduan_deploy.
  • TestedOn a real Ubuntu 24.04 server: CasaOS, nginx, and Apache were moved and restored; a simulated failure rolled back cleanly; a site from Saka got a Let's Encrypt certificate while CasaOS stayed reachable by IP; releasing the server restored CasaOS to port 80.
Oct 62026
feature2 updates#

Guide: share an app built with AI with your clients

A new guide for apps built with OpenCode, Claude Code, or Cursor that only run on 127.0.0.1: a tunnel preview from your own computer, always online from Git, or a static site.

Docs

  • From an empty VM to online, by AINew panduan_deploy MCP tool and a docs section: an AI agent (OpenCode, Claude Code, Cursor) creates or connects an Ubuntu server, pushes the code to Git, deploys it with an HTTPS address (a sslip.io address when there is no domain), fills secrets from the local .env, checks it works, and sets up pull request previews for client revisions. Includes the app rules that trip AI-built apps: no persistent disk (SQLite and uploads are lost on deploy), PORT and 0.0.0.0, production build, and no secrets in chat.
  • Share an app with clientsWhy 127.0.0.1:5173 cannot be shared, the Vite allowedHosts setting tunnels need, serving the build instead of dev mode, deploying from Git with pull request previews for client revisions, and connecting Saka to OpenCode over MCP so it can do the steps for you.
Oct 52026
fix2 updates#

Important alerts reach you by email too

Users who have not linked Telegram now get every important alert by email instead of missing it.

Notifications

  • Email when Telegram is not linkedServer down, disk almost full, high RAM/CPU, stopped apps, site error spikes, failed cron jobs, failed backups, SSL problems, and load balancer health now fall back to your verified email when Telegram is not linked. Before, only some alerts did and the rest were lost.
  • No noiseLight notices such as agent updates stay Telegram only; each problem is still sent once, with one recovery message.
Oct 42026
featuresecurity4 updates#

Saka Shield: rate limits, country blocks, and an OWASP WAF

Shield can now cap requests per IP, block chosen countries, and run the Coraza OWASP Core Rule Set in monitor or block mode.

Security

  • Rate limit per IPIPs sending more than 600 requests a minute (adjustable 300 to 3000, or off) are blocked temporarily.
  • Country blocksCountry IP ranges are loaded into the server firewall (web ports by default, all ports optional), and Cloudflare visitors are blocked by Cloudflare's country header.
  • OWASP WAFCoraza with the OWASP Core Rule Set for every site on the server. Monitor mode lists what would be blocked (attack type, IP, address, rule); block mode answers 403 and blocks repeat attackers in the firewall. About 50 MB of RAM, flat. The WAF build is downloaded only when turned on and verified against a hash inside the signed agent.
  • Tested900 fast requests were cut off after 600 per minute; a Singapore block closed the web for a Singapore VM while SSH stayed reachable; block mode rejected SQL injection, XSS, and path traversal while normal pages and searches passed.
Oct 42026
feature4 updates#

App error alerts and scheduled job monitoring

Get told when an app starts throwing server errors, when a Git app's cron job fails, or when any scheduled job stops reporting.

Monitoring

  • Error spikes, with the logWhen a site answers 5xx for at least 10 requests and 20% of traffic within 5 minutes, you get a Telegram alert with the most affected addresses and the latest error lines from the app (PHP fatal errors, tracebacks, panics), plus a warning on the site page. Works for any app; at most hourly per site, with a back-to-normal message.
  • Git cron failuresScheduled cron jobs of Git apps alert you once when they start failing (with the end of their output) and again when they succeed.
  • Ping addresses for any jobSettings, Scheduled job monitoring: give a job a name and interval, then have it call its ping address after each run (or /gagal on failure). Late or failed jobs are reported, and recoveries too. MCP tools tugas_pantau_daftar, tugas_pantau_buat, tugas_pantau_hapus.
  • TestedOn a real server: a PHP page failing with a database error raised an alert with the PHP fatal error lines; a failing cron job alerted once and reported recovery after the fix; a ping monitor went through new, ok, failed, late, and ok again.
Oct 42026
featuresecurity6 updates#

Saka Shield blocks attackers at the server level

A server-level guard for any app: IPs that brute-force SSH, scan for weaknesses, flood missing pages, or guess login passwords are blocked in the server firewall.

Security

  • Any app, not just WordPressThe agent reads SSH and web access logs every 20 seconds and blocks offenders in nftables (every port and Docker container) and in Caddy (for visitors behind a load balancer, tunnel, or Cloudflare). Blocks expire on their own; repeat offenders are blocked for 24 hours.
  • On by default for new serversServers added from now on start with Shield on, excluding the IP that created the install command; existing servers keep it off until you turn it on.
  • Never locks you outYour IP when you turn it on, connected SSH sessions, private IPs, Cloudflare ranges, the server itself, and load balancer nodes are never blocked. Add exceptions and lift blocks from the server page.
  • Real visitor IPs behind CloudflareSites now trust Cloudflare's forwarded headers, so logs, X-Real-IP, and Shield see the real visitor instead of a Cloudflare address.
  • TestedAn attacking VM was blocked on every port within 45 seconds after 8 SSH attempts, and again after probing 8 weakness paths; turning Shield off removed every rule.
  • For AI agentsMCP tools perisai_status, perisai_atur (needs approval), perisai_lepas.
Oct 42026
featuresecurity4 updates#

Saka Shield: hacked-file detection and a file shield for every site

The daily WordPress check now spots files changed by intruders and can restore the originals, and every site hides files that must never be public.

Security

  • File integrityWordPress core and wordpress.org plugin files are compared with their original checksums. Changed files and unknown files in wp-admin or wp-includes are flagged as important; one button re-downloads the original core or plugin files (same version, backup first, automatic restore if the site breaks).
  • File shieldWordPress, PHP, and static sites answer 404 for .git, .env, .sql dumps, wp-config.php backups, debug.log, and error_log, in front of PHP.
  • TestedOn a real server: a changed core file, a backdoor in wp-includes, and a changed plugin file were all detected; reinstalling restored the originals and only the backdoor remained listed for the owner to remove.
  • For AI agentsNew MCP tool wp_pasang_ulang (needs approval).
Oct 42026
featuresecurity5 updates#

WordPress staging and daily security checks

Try plugin and theme updates on a staging copy, push them to the live site safely, and get alerted about vulnerable plugins.

WordPress

  • Daily security checkWordPress, plugin, and theme versions are checked against wordpress.org every day. Plugins closed for a security issue and missed WordPress fix releases are flagged as important and reported on Telegram; outdated and abandoned plugins are listed too. Important plugins can be deactivated from the panel.
  • Update all, safelyOne button updates WordPress, plugins, themes, and languages with a backup first and automatic restore if the site breaks.
  • StagingCreate a copy at staging-<id>.<ip>.sslip.io with its own database, HTTPS, no indexing, and outgoing email and WP-Cron turned off. Update or change it, then push files only (live database, uploads, and wp-config untouched) or files and database, with a backup and automatic restore.
  • TestedOn a real server: a plugin closed for security was flagged, updates were tried on staging while the live site stayed unchanged, a files-only push kept a post created on the live site in the meantime, and a full push switched addresses back to the live domain.
  • For AI agentsMCP tools wp_keamanan, wp_perbarui, wp_nonaktifkan_plugin, wp_staging; actions on the live site need your approval.
Oct 42026
featurefix5 updates#

Snapshot, resize, and delete servers in your cloud account

Servers at DigitalOcean, Vultr, and IDCloudHost can now be snapshotted, resized, and deleted at the provider straight from the server page, so no forgotten VM keeps billing you.

Servers

  • In your cloud accountA new card on the server page shows the VM's provider, region, size, and snapshots. Servers not created through Saka are matched automatically (by public IP, or hostname on IDCloudHost).
  • Snapshot and resizeTake a snapshot while the server runs; resize to mini, small, medium, or large with an optional snapshot first (about 1-3 minutes of downtime). Tested on DigitalOcean and IDCloudHost.
  • Delete at the providerDelete the VM and remove it from Saka in one step after typing the server name. On IDCloudHost the reserved public IP is released too, which otherwise kept being billed.
  • For AI agentsMCP tools server_cloud, server_snapshot, server_ubah_ukuran (needs approval), server_hapus_cloud (needs confirmation).

Fixes

  • Cloud server listServers that were already removed no longer show up as connected in the list of servers created through your cloud account.
Oct 42026
featurefix5 updates#

Tunnel and load balancer monitoring, public status pages

Saka now watches tunnels and load balancers from the internet like it already did for sites, and you can share a public status page with your customers.

Monitoring

  • Tunnels and load balancersEvery active <name>-app.hack.id tunnel and every load balancer domain is opened from the internet each minute. The Tunnel and load balancer pages show whether it is reachable, and you get a Telegram alert after 3 minutes down and when it recovers.
  • Public status pageSettings, Public status page: pick what to show and give each a display name, then share app.saka.work/status/<code>. It shows the current state and 30 days of history per service, without domains, servers, or error messages. JSON at the same address plus .json; MCP tool pantau_daftar.

Fixes

  • Point-in-time restore of the last few minutesPostgreSQL clusters now ship WAL to S3 at least every minute, so a restore to a few minutes ago works even on quiet databases. Existing clusters with backups are fixed automatically. If the chosen time has not reached S3 yet, the message says to wait a minute or pick an earlier time.
  • Turning on backupsTurning on PostgreSQL backups could fail with "not the primary node" when the primary moved while containers were recreated; Saka now asks again for the current primary.
  • Pull request databases on MariaDB/MySQLFixed a quoting bug that stopped the per-PR database copy on MariaDB and MySQL. Tested end to end: a signed pull request webhook builds the preview with its own database, a new push reuses it, and closing the PR removes both; production stayed untouched.
Oct 42026
feature4 updates#

Start here guide, bring tables back from a backup, and a database per pull request

New users get a step-by-step guide for their goal, restored copies can hand tables back to the live database, and every pull request preview gets its own database.

Panel

  • Start herePick what you want to do (move from cPanel, install an AI assistant, survive a region outage, Git apps, or run it through an AI agent) and the Servers page shows the steps with check marks that tick themselves as you go, each with a link to the right page and guide. Reopen it any time from the footer.
  • Clearer namesFleets are now called Bulk deploy, with an explanation and a guide link on the page. Request a feature moved to the footer.

Databases

  • Bring data back from a copyA restored copy now lists every table with its row count in the copy and in the database now, marking the ones that differ. Copy the chosen tables into a separate pulih_<time> place to pick rows with SQL, or replace the current contents after they are saved to sebelum_<time>. Works for PostgreSQL and MariaDB; MCP tools database_salinan_tabel and database_salinan_ambil.

Git apps

  • A database for every pull requestPull request previews now get their own copy of the app database (<name>_pr<number>, from DATABASE_URL or Laravel's DB_DATABASE) on the same database server, so migrations and test data never touch production. Reused on later pushes and removed with the preview. If the copy cannot be made, the preview does not run instead of quietly using production.
Oct 32026
feature4 updates#

Bulk deploy: one app to many servers, safely

Group the same Git app across many servers and deploy it with a canary, pinned commit, and automatic stop on failures.

Bulk deploy

  • Staged deploysOne canary first; if it succeeds, the same commit goes to the rest 5 at a time. The rollout halts on its own if the canary fails or more than 10% fail, leaving untouched servers as they were.
  • Shared variablesChange environment variables on every member at once, in waves with a canary.
  • See every versionThe Bulk deploy page shows which commit runs where; disconnected servers are skipped, not failed. One Telegram summary instead of a message per server.
  • For AI agentsMCP tools armada_daftar, armada_buat, armada_status, armada_deploy, armada_env.
Oct 32026
feature2 updates#

Background jobs follow your app to the second location

Workers and cron of a two-location Git app now move to the second server when the main one is down, and come back without running twice.

Git apps

  • TakeoverIf the main server is unreachable for 3 minutes, the second server starts the workers and runs cron; it hands them back 2 minutes after the main server is reachable again.
  • Never twiceA main server that has just booted holds its workers and cron for 4 minutes until the handback; in testing there was no overlap at all.
Oct 32026
feature5 updates#

Database backups and point-in-time restore

PostgreSQL and MariaDB clusters send every change to your own S3 bucket and can be restored to any second, into a separate copy so the running database is never touched.

Databases

  • Continuous backups to your S3pgBackRest streams WAL continuously plus weekly full and daily differential backups to Amazon S3, R2, Spaces, Wasabi, or any compatible bucket. Backups are encrypted with a recovery code only you and your servers hold.
  • Restore to any second, safelyPick a time and Saka builds a separate copy of the database as it was then, on one of your data servers. Check it, take what you need, then delete it; the live cluster keeps running untouched.
  • MariaDB tooDaily mariadb-backup plus binlogs shipped every minute (restore point at most 1-2 minutes behind). If the backup node dies, another node takes over and makes its own full backup right away.
  • TestedOn PostgreSQL 18 and MariaDB 10.11, 150 deliberately deleted transactions came back intact from a copy taken just before the deletion; the MariaDB backup made in Singapore was restored in Bangalore.

Servers

  • Create servers from your cloud accountConnect DigitalOcean, Vultr, or IDCloudHost and create servers from the Add server dialog or through your AI agent; the Saka agent installs itself and the server appears within minutes.
Oct 22026
featurefix5 updates#

Add Telegram later, steadier access, and an HA guide for AI agents

Hermes and OpenClaw can get a Telegram bot after installation, the panel loads more reliably from Indonesia, and AI agents get a step-by-step HA guide.

Apps

  • Telegram after installationSkipped the bot token when installing Hermes or OpenClaw? Add it, or change it, on the app's Telegram card without reinstalling. Data is kept and the app is put back as it was if anything fails. Over MCP: aplikasi_telegram.
  • Model name suggestionsA mistyped AI model name now fails with the closest names the provider actually offers, instead of a generic error.

Panel

  • Steadier access from Indonesiasaka.work and app.saka.work now reach the panel through a Cloudflare Tunnel, fixing 522 errors some visitors in Indonesia saw.
  • Windows via WSLThe getting started guide explains installing the agent on Ubuntu in WSL 2, including turning on systemd.

For AI agents

  • HA guide over MCPThe panduan_ha tool gives an AI agent the order of steps and rules for keeping an app and its database running when one VM or region goes down.
Oct 12026
featurefix7 updates#

MySQL apps that survive a region outage

Apps on MySQL 8 can now run on two VMs in two regions with MariaDB, and one small witness can serve many clusters.

Databases

  • Move from MySQL 8The cluster page copies an existing MySQL 8 or MariaDB database into a MariaDB cluster: collations mapped, MyISAM converted to InnoDB, row counts compared table by table, tables without a primary key reported.
  • MariaDB 10.11 for older appsOld MySQL client libraries (e.g. Ruby 2.5 with mysql2) read every text value as 0 from MariaDB 11.4 and newer. Clusters can now use MariaDB 10.11 LTS, which works with them.
  • Web and database on the same VMA MariaDB cluster proxy can run on a cluster member: it checks each node's health and always leads to the same writer.
  • Shared witnessOne server can be the witness for many clusters, so each customer needs two VMs plus one shared small witness instead of three VMs.
  • Version shown correctlyThe panel now shows the real database version; PostgreSQL 18 clusters were labeled 17.

Apps

  • Clearer start errorsWhen an app stops right after starting, the error now hides stack-trace lines so the line naming the cause is visible.

Tested

  • 13 seconds, nothing lostA Ruby 2.5 app on MariaDB 10.11 with data from MySQL 8, Singapore and Bangalore: the main VM was powered off hard and the second region served again in about 13 seconds with zero lost transactions.
Sep 302026
featurefix6 updates#

Web app and database that survive a VM outage

Run an app and its PostgreSQL on two VMs that take over for each other, and move an existing database into a cluster with one button.

Databases

  • Move an existing databaseThe cluster page copies an old PostgreSQL (e.g. on the app's own VM) into the cluster in a single transaction, then compares row counts table by table. A failure leaves the cluster unchanged and the old database is never modified.
  • Web and database on the same VMA PostgreSQL cluster proxy can now run on a cluster member. It checks Patroni and always leads to the primary, never the local replica, and Git containers reach it at host.docker.internal.
  • Clear limitsA server that already belongs to one cluster is refused with a clear message instead of failing midway.

Git apps

  • HTTPS on the second locationThe second location of a Git app now gets its own certificate; before, its certificate challenge was answered by the app on the main server.
  • Simple Git serversRepos on servers without shallow clone support (plain HTTP) deploy, and the second location builds the exact commit of the main one.

Tested

  • 21 seconds, nothing lostWith the main VM holding both the app and the primary database powered off hard, the second VM took over in about 21 seconds with zero lost transactions in synchronous mode. A step-by-step guide is in the docs.
Sep 292026
feature4 updates#

Sites monitored from the internet

Saka now opens every site from the outside each minute and tells you when visitors can't reach it, and why.

Sites

  • Checked like a visitorEach site's main domain is opened over HTTPS every minute; the site page shows reachability, response time, and 30-day uptime.
  • Alerts with the causeAfter 3 minutes of failures you're told on Telegram (or email) whether it's DNS, the HTTPS certificate, a closed port, a timeout, or a 5xx error, and again when it recovers.
  • Quiet until it worksSites whose DNS isn't pointed yet don't raise alerts, and a server that's disconnected is reported once, not per site.
  • Renewal watchAn automatic certificate with under 7 days left means renewal failed, and you're told before visitors see a warning.
Sep 292026
featuresecurity4 updates#

Invite your team

Let colleagues or clients manage servers in your account with their own Saka login, limited to the role and servers you choose.

Team

  • Invite by emailSettings, Team: enter an email, pick Viewer, Developer, or Admin, and choose all servers or specific ones. Invitees sign in with their own account; no shared passwords.
  • Switch accountsMembers switch between their own account and the accounts they manage from the account menu; a banner shows whose account and which role.
  • Limits that holdServer limits apply to servers, sites, apps, tunnels, databases, load balancers, tasks, and alerts. The terminal, API tokens, AI connectors, and account settings stay owner-only.
  • TraceableThe task history records which member ran each task, and the owner is told on Telegram when a member joins or deletes something. Removing a member ends access immediately.
Sep 292026
feature3 updates#

Use your own HTTPS certificate

Install a paid, wildcard, or EV certificate on a site, or keep the free automatic one (suggested by a user).

Sites

  • Paste or uploadThe HTTPS card takes a PEM certificate with its CA bundle and the private key; the key stays on the site's server.
  • Checked before useKey pair, validity, intermediate chain, and covered domains are checked; domains the certificate doesn't cover keep Let's Encrypt.
  • Expiry remindersTelegram reminders 30, 14, 7, 3, and 1 days before the certificate ends, and a one-click return to automatic certificates.
Sep 292026
featuresecurity4 updates#

Tunnels to other computers on your network

Open a web app running on another computer on the same network, such as XAMPP on a Windows PC, through a free Saka address.

Tunnels

  • Another device on the networkBesides a port on the Saka server, a tunnel can point to a local IP (10.x, 172.16-31.x, 192.168.x) such as 10.10.104.240:80.
  • Admin pages stay privateTunnels to web apps block /phpmyadmin, /xampp, /webalizer, /adminer, and server-status from the internet, since XAMPP's phpMyAdmin often has no password.

Servers

  • Rename a serverRename servers from the server page (suggested by a user).
  • Small serversServers under 1 GB of RAM are flagged when creating a site instead of failing later, and install results reach you by email if Telegram isn't linked.
Sep 292026
feature2 updates#

Connect Saka to claude.ai and ChatGPT

Add Saka as a custom connector in claude.ai, Claude Desktop, or ChatGPT with a sign-in instead of a copied token.

AI & MCP

  • OAuth sign-inEnter https://app.saka.work/mcp as a connector address; Saka asks you to sign in and choose View only or Can make changes, then sends you back.
  • Stays under your controlEach connector is an API token you can see and revoke in Settings, and you're told on Telegram when one connects.
Sep 292026
feature4 updates#

Git apps in two locations

Apps deployed from Git can now run on a second server too, following every release of the main one.

Git apps

  • Same commit, same variablesThe second server builds the release that runs on the main server; env variables and the deploy key are sent directly between the servers, never through Saka.
  • Follows every changeDeploys, rollbacks, variable changes, and worker/cron settings on the main server are applied to the second one; if that fails you are told and the main server is unaffected.
  • No double jobsWorkers and cron run only on the main server.
  • Automatic failoverA DNS load balancer is created as for other sites: visitors go to the second server when the main one is down.
Sep 292026
featurefix8 updates#

Servers that look after their own memory

Saka now shows which app is eating RAM, warns before a server stalls, and fixes the most common cause by itself.

Servers

  • Usage per appThe server page lists RAM and CPU for each app and site, largest first, updated every minute.
  • High RAM and CPU alertsA Telegram message when RAM stays above 90% for 5 minutes or CPU above 90% for 10 minutes, naming the biggest users and what to do.
  • Automatic swapWhen RAM is nearly gone and there is no swap, or an AI assistant runs on a server under 4 GB, the agent creates a 2 GB swap file and tells you.
  • Self-recovering agentAn agent that can't reach Saka for 3 minutes restarts itself, and it is the last thing the kernel stops when memory runs out. Heartbeats no longer wait on slow status checks.
  • Clearer alertsBrief blips while an app restarts or updates are not reported; a disconnect caused by low memory says so; times are shown in WIB.

AI assistants

  • Uses Saka tools, not the shellConnecting adds a short note to the assistant's instructions so it checks the server through Saka instead of running commands inside its own container. Removed on disconnect.
  • Open portsNew MCP tool server_port lists open ports and their programs.

Account

  • One verification emailSigning up sends a single code, valid for 10 minutes; resends are limited to one per minute.
Sep 292026
feature9 updates#

Manage servers by chat, and a safer sign-in

Your OpenClaw or Hermes assistant can now look after your servers through Saka, sign-in asks for an email code, and routine server jobs no longer need the terminal.

AI assistants

  • Connect to SakaOne button on the OpenClaw or Hermes page: ask on Telegram how your servers are doing, read logs, or run maintenance. The token lives only on the app's server and is revoked when you disconnect or delete the app.
  • View-only tokensTokens (for assistants, MCP, and the API) can be view only: tools that change things are hidden and refused. With change access, every change an assistant makes is reported on Telegram.
  • Local AI gateways9Router, LiteLLM, or Ollama on the same server can be used as the AI provider with http://localhost addresses; Saka points the app at the server itself and explains what to fix when the gateway only listens on 127.0.0.1.

Account

  • Email code at sign-inAfter the correct password, a 6-digit code is sent to your email (it names the IP that signed in). Tick "remember this device" to skip it for 30 days; changing your password forgets all devices.

Servers

  • Maintenance cardUpdate system packages, clean up disk (rollback images and data are kept), reboot with confirmation, and read the system log, all from the server page.
  • Reboot when neededThe "waiting for reboot" readiness check now has a Fix button and is also detected on AlmaLinux and Rocky.

Load balancers

  • Backup pool in proxy modeBackup servers now actually take over when every main target is down (about half a second), and traffic returns to the main servers when they recover.

Fixes

  • Files owned by the right userSettings files written for apps are now owned by the app's user instead of root.
  • Clearer terminal password promptThe emergency terminal asks for your Saka account password and says so, not the server's password or SSH key.
Sep 292026
feature5 updates#

The saka CLI and an emergency terminal

Manage Saka from your own terminal or CI, and open a shell on a server from the panel when there is no other way.

CLI

  • One-line installcurl -fsSL https://saka.work/cli.sh | sh on Linux and macOS, or irm https://saka.work/cli.ps1 | iex on Windows; the binary is checked against its SHA-256.
  • Everything from the terminalServers, sites, deploys, rollbacks, variables, backups and downloads, cron jobs, and any API endpoint, with --json output and exit codes for scripts.
  • Deploy from CISet SAKA_TOKEN and run saka site deploy; a GitHub Actions example is in the docs.

Servers

  • Emergency terminalA root shell on the server from the panel, through the agent, with no open port or SSH key. Asks for your password each time, closes after 15 idle minutes, and notifies you on Telegram.
  • Owner stays in controlOnly the account owner can open it (not API tokens, MCP, or Saka support); the session is not recorded; the server owner can turn it off with "terminal": false.
Sep 282026
feature9 updates#

Sites in two locations, and servers that aren't standard

A site can now run on two servers and stay up when one goes down. Installs also cope better with VMs that aren't set up like a typical cloud image, and explain clearly when the server's network is the problem.

Two locations

  • Second locationCopy a WordPress, PHP, or static site to another server with one click; files are copied every minute and the database is shared through a MariaDB cluster.
  • Automatic failoverA DNS load balancer is created for you: visitors go to the main server, and to the second one within seconds if the main server is down. HTTPS works on both.
  • Nothing lost on returnFiles changed on the second server while the main one was down are pulled back before copying resumes.
  • Locked-down copyingThe copy key can only sync that one site folder, only from the main server, with no shell.

Installs

  • Try again without retypingA failed or interrupted app install can be retried from the panel; the details, including keys, are kept only on your server.
  • Clear network errorsIf the server reaches Saka but its IP can't reach package or image registries, the error says so and suggests changing the IP or contacting the provider.
  • More kinds of VMsUbuntu without the universe component, apt locked by updates, half-finished dpkg runs, and one-click clock sync are handled automatically.

Fixes

  • Tasks survive restartsA running install is no longer cancelled, or left stuck as running, when Saka Panel restarts during an update.
  • Chart range menuThe CPU chart range menu no longer closes by itself while the page refreshes.
Sep 282026
feature6 updates#

Workers, cron, and pull request previews

Git apps can now run background workers and scheduled jobs, and every pull request gets its own preview address. Backups can be downloaded from the panel.

Git apps

  • WorkersLong-running processes without a port (queues, bots) from the same release and variables, like a Procfile. They follow deploys and rollbacks, and a worker that crashes right away is reported.
  • Cron jobsReady-made or 5-field schedules in the server's time zone. A job never overlaps itself, is stopped at its time limit, and its last exit code and log are shown, with Run now.
  • Pull request previewsEach pull request (GitHub, Gitea, GitLab, forks included) is built on your server and served at its own HTTPS address, updated on every push and removed when the PR closes. SAKA_PRATINJAU=1 lets your code pick a test database.
  • For AI agentsMCP tools situs_proses, situs_cron_jalankan, and situs_pratinjau.

Backups

  • Download from the panelSave any backup as a .tar.gz straight from your server; nothing is stored on Saka. Backups that exist only in S3 can be fetched to the server first.

Fixes

  • MCPAI agents connecting over MCP were rejected with 403 behind the proxy; they connect normally again. Every release is now checked end to end (sign in, panel, API, MCP) and rolled back automatically if something breaks.
Sep 282026
feature6 updates#

Backups to your own S3

Site backups can now go to your own S3 storage, encrypted, and be restored to a brand new server if the old one is lost.

Backups

  • Any S3Amazon S3, Cloudflare R2, DigitalOcean Spaces, Wasabi, Backblaze B2, or any S3-compatible service. Access is tested (write, read, delete) before saving.
  • Encrypted before uploadBackups are encrypted on your server with age. The S3 keys and the encryption key never leave your server, so neither the provider nor Saka can read them.
  • One recovery codeA single line holds the destination and the key. Paste it on a new server to see and restore every site backed up to that bucket.
  • Restore anywhereRestore a site in place from S3, or recreate it on another server with a new domain; WordPress addresses and ready-made app databases come along.
  • Hands-offUploads run in the background with a retention you choose, and you get a Telegram message if sending keeps failing.

Fixes

  • Sign inSigning in no longer asks for a country; for a few hours on Sep 28 some sign-ins were wrongly rejected.
Sep 282026
feature4 updates#

Ready-made apps with their own database

Apps that need PostgreSQL or MySQL now install with one click too, database included.

Catalog

  • Three new appsUmami (website analytics), Ghost (blog and newsletter), and Listmonk (newsletters and bulk email).
  • A database per appEach app gets its own PostgreSQL or MySQL on a private network, never exposed outside the server. Nothing to set up.
  • One package for backupsThe database is part of the app's daily backup, update, and restore, so a failed update rolls back both the app and its data.
  • No default passwordsApps that ship with a well-known admin password get a random one at install; see it with Show login on the app page.
Sep 282026
feature5 updates#

Ready-made apps

Install popular self-hosted apps with one click, each with a domain, HTTPS, daily backups, and safe updates.

Catalog

  • Eight apps to startn8n, Uptime Kuma, Vaultwarden, Metabase, NocoDB, Gitea, Memos, and Nextcloud. Servers without enough RAM for an app are warned before installing.
  • Data you keepApp data lives on your server and is included in daily backups; the app pauses briefly during a backup so the data is consistent. Restore from the Backups card.
  • Safe updatesUpdate to the latest version from the app page: a backup is made first, and the previous version and data come back automatically if the new version fails.
  • Settings that can't break the appIf a changed variable stops the app from starting, the previous variables are restored automatically.
  • For AI agentsMCP tools katalog_daftar and situs_perbarui_aplikasi; install with situs_buat type katalog.
Sep 282026
feature7 updates#

Apps from Git

Deploy your own app straight from a Git repository to your own server: push to deploy, zero downtime, and one-click rollback.

Git apps

  • Any stackA Dockerfile is used if present; otherwise Railpack detects Node, Python, Go, PHP, Ruby and more. Monorepo subfolders and custom build/start commands are supported.
  • Zero-downtime deploysEach release starts next to the old one and must pass a health check before visitors switch. Failed builds or crashes leave the old release serving, and you are notified on Telegram.
  • One-click rollbackThe last five releases are kept as images; rolling back takes seconds, no rebuild.
  • Private repos without account accessThe server creates a read-only deploy key; Saka never asks for your GitHub token. The code never passes through Saka.
  • Deploy on pushSigned webhooks from GitHub, GitLab, or Gitea; pushes during a deploy are queued.
  • Environment variablesStored only on the server, values never shown again, applied with a zero-downtime restart.
  • For AI agentsMCP tools situs_deploy, situs_rollback, and situs_env, plus REST endpoints.
Sep 282026
feature17 updates#

Home servers and free tunnel addresses

Run Saka on a computer at home instead of renting a VPS. Sites and web apps get a free public address with HTTPS, even behind a home router or CGNAT.

Tunnels

  • Free <name>-app.hack.id addressesPick a name, choose a Saka site or a local port, and Saka sets up a Cloudflare Tunnel on your server. No router ports to open, and your home IP stays hidden.
  • Use the tunnel as the site's main addressWordPress links switch to the tunnel address automatically, and switch back when the tunnel is deleted.
  • Clean removalDeleting a tunnel, a site, or releasing a server removes the connector, DNS record, and tunnel.
  • For AI agentsNew MCP tools tunnel_buat, tunnel_daftar, and tunnel_hapus, plus the /api/v1/tunnel REST endpoints.

Hermes

  • Already running Hermes? Saka takes it overSaka detects a Hermes you installed yourself (Docker or directly) and refuses to install a second one. Manage with Saka stops the old one without deleting it, copies its memory, skills, sessions, bot, and keys, and restores it automatically if anything fails.
  • Manage Hermes from the panelApprove people who message your Hermes bot, add or remove Telegram IDs, switch AI provider, model, or key without reinstalling, check for updates with an automatic backup first, and make backups.

Guides

  • Home server guideStep by step for a mini PC or old laptop with Ubuntu, Raspberry Pi with Ubuntu Server, Windows via WSL2 (beta), and Mac via Multipass (beta).

Improvements

  • Feature requests pageTwo column layout: the form stays in view while you browse and vote on other requests.
  • Faster supportWhen you report a problem, the Saka team can now check your apps and sites, open your panel view only for 30 minutes, and run safe fixes such as restarting an app. Every action is logged and announced to you on Telegram with the reason.

Databases

  • PostgreSQL 18New database clusters use PostgreSQL 18 by default, with 17 still available. Tested with the primary forcibly powered off: failover in 29 seconds without data loss.

Mobile and tablet

  • Comfortable on phones and iPadsA menu drawer instead of a cut-off menu row, lists shown as cards on phones, bigger touch targets, dialogs as bottom sheets, and no page zoom when typing on iPhone.

Accounts

  • Email verification with a codeNew accounts confirm their email with a 6-digit code before using the panel. Existing API and MCP tokens keep working.
  • Forgot passwordReset your password with a one-time link by email, valid for 1 hour. All other sessions are signed out.
  • Emails in your languageAccounts from Indonesia get emails in Indonesian, everyone else in English.

Safety fixes

  • Releasing a server in use is blockedA server that is still a balancer or target of a load balancer, or a node of a database cluster, can't be released until that is removed first. The panel lists what still uses it.
  • Deleting a cluster in use is blockedA database cluster that is still the database of a site can't be deleted.
  • Updates never interrupt running workAutomatic agent updates now wait until tasks such as releasing a server have finished.
Sep 282026
featuresecurity5 updates#

Signed updates, feature requests, and load balancer firewall

Agent updates are now cryptographically signed, you can ask for features directly from the panel, and load balancers can allow or block IP ranges.

Security

  • Signed agent releasesEvery agent update is signed with Ed25519. The agent refuses any update without a valid signature, so even a compromised control plane cannot push code to your server.
  • Security pagesaka.work explains exactly what the agent does on your VM, what it never does, and what data is sent.

Features

  • Feature requestsSuggest features, vote on others, and get a Telegram message when the status changes. Your AI agent can file requests through MCP after you approve.
  • Load balancer firewallAllow or deny IPs and CIDR ranges per load balancer, editable after creation.
  • Account menuSettings and sign out now live under your email in the top bar.
Sep 282026
feature7 updates#

Sites: a cPanel replacement for app owners

Host WordPress, PHP, or static sites on your own server with the essentials people used cPanel for, and move in from cPanel or another VPS.

Sites

  • WordPress, PHP, and static sitesAutomatic HTTPS, jailed SFTP per site, local or clustered database, daily backups with one-click restore.
  • phpMyAdmin per siteOpens through a secure dashboard link with no open ports, limited to that site's database, and turns itself off after 3 hours.
  • PHP settingsUpload limit, memory, execution time, max_input_vars, time zone, and error display, without editing php.ini.

Migration

  • Move from cPanel with a loginLike WHM's Transfer Tool: Saka triggers a full cPanel backup, downloads it, and imports it. API tokens are supported for accounts with 2FA.
  • Move between serversCopy a site directly from one server to another over one-time encrypted transfer, keep the same database credentials and SFTP account, and optionally update WordPress with automatic rollback.

AI assistants

  • Google GeminiGemini keys work for OpenClaw and Hermes, validated before install.
  • Hermes fixesUses the published image tag and always sets the AI provider.
Sep 282026
featurefix7 updates#

English, reliability, and mobile

Saka Panel and saka.work are now available in English, and the control plane runs in two locations.

Languages

  • English everywhereThe panel defaults to English, with Indonesian one click away. Server messages and API errors follow your language.
  • Sign up with country and phonePhone verification by OTP will follow.

Reliability

  • Two location control planeA warm standby takes over if the primary goes down. Apps on your servers keep running either way.
  • Recovery after restartsDatabases, load balancers, and sites that were being created during a restart are finished or marked failed automatically.
  • IP changesWhen a server's IP changes, you are notified and database clusters and load balancers are updated.

Fixes

  • Mobile layoutPanel menu, comparison tables, and docs fit small screens.
  • Already signed inOpening the sign in page while signed in now goes straight to the panel.
Sep 282026
feature4 updates#

Load balancers

Spread visitors across servers with health checks and automatic HTTPS, running on your own servers with no separate load balancer bill.

Load balancers

  • One or two balancersCaddy based, health checked every 5 seconds, automatic Let's Encrypt certificates, round robin, least connections, or sticky sessions.
  • Health aware DNSEach load balancer gets DNS that only answers with healthy servers, so mobile apps, APIs, and webhooks avoid dead servers too. A DNS only mode covers non web services.
  • Weights and regionsSend more traffic to bigger servers, or send visitors to the nearest region.

Policy

  • Saka never stops your appsThe earlier idea of pausing apps after a long disconnect is gone. If Saka is unreachable or the agent is removed, everything on your server keeps running.
Sep 272026
feature4 updates#

Managed databases across locations

Run PostgreSQL or MariaDB as a cluster across servers in different locations, with automatic failover.

Databases

  • PostgreSQL 17 with PatroniReplicas in other locations over WireGuard, automatic promotion when the primary fails, replica lag in the panel.
  • MariaDB with GaleraMulti primary clusters with an arbitrator, plus a proxy on your app server so apps always reach a healthy node.
  • FencingA failed node is fenced before a new primary takes over, to avoid split brain.

Operating systems

  • More Linux distributionsUbuntu 22.04, 24.04, 26.04, Debian 13, AlmaLinux and Rocky Linux 8 to 10, and CentOS Stream 9 and 10.
Sep 272026
feature7 updates#

First public release

Connect any VPS with one command and manage it from the panel or through your AI agent, without giving the AI root access.

Core

  • Agent and panelA small agent connects out from your server; nothing listens on a new public port.
  • MCP and REST APILet Claude, ChatGPT, or other agents check and manage your servers through safe, confirmable tools.
  • Server readinessSeven checks with plain language fixes, including one-click swap and a firewall that keeps the ports you use open.
  • Release serverOne clean way out that removes everything Saka installed and leaves what was there before.

AI assistants

  • OpenClaw and HermesOne-click install with Telegram, WhatsApp for OpenClaw, and any OpenAI compatible provider.
  • Manage OpenClaw without the terminalAccess requests, AI models, skills, schedules and reminders, memory, persona, backups, and safe updates with automatic rollback.
  • App dashboardsOpen an app's own dashboard through a private link with no open ports.

Not included yet