Getting started
From creating an account to seeing your first server in the panel, usually in under five minutes.
1. Create an account
- Open app.saka.work/masuk/?daftar=1.
- Enter your email and a password (at least 10 characters).
- Press Buat akun (Create account).
- Open your email and enter the 6-digit code from Saka (valid for 15 minutes). The panel opens once your email is verified. Mistyped your email? Choose Wrong email? to fix it.
People signing up from Indonesia get emails in Indonesian, everyone else in English. Forgot your password? Choose Forgot password? on the sign in page; the reset link works for 1 hour and only once.
Your servers stay yours. If Saka Panel cannot be reached or the agent is removed, your apps and databases keep running. To leave cleanly, see Release a server.
2. Add a server
Get a VPS from any provider. Requirements:
- A supported OS (see the list below), with systemd.
- root access, or a user with sudo.
- At least 512 MB of RAM for the agent alone. 2 GB for OpenClaw, 1 GB for Hermes.
- At least 1 GB of free disk space for the agent. OpenClaw needs 8 GB free.
- The server can reach the internet over HTTPS (outbound).
- x86_64 (amd64) or ARM64 architecture.
In the panel, open Server, press + Tambah server (Add server), give it a name, then press Lanjut: buat perintah pasang (Next: create install command). You get a one-line command like this:
curl -fsSL https://app.saka.work/pasang.sh | sudo sh -s -- <token-daftar>Open a terminal on the server (with ssh root@IP-SERVER or your provider's web console), paste the command, and press Enter. The panel page updates by itself as soon as the server connects.
What the install script does, in order:
- Checks the OS, architecture, RAM, free disk space, and the connection to Saka Panel. If anything is missing, the script stops and explains how to fix it without changing anything.
- Downloads the agent and verifies its SHA-256 checksum. Corrupted or modified downloads are rejected.
- Registers the server to your account using the token.
- Installs the
saka-agentsystemd service and starts it.
The registration token is valid for 24 hours and can be used only once. If it expires, create a new command from the panel. The script is safe to run again: if the server is already registered, it stops without changing anything.
The agent updates itself automatically when the Saka Panel version changes: it downloads, checks the release signature and SHA-256, tests, then swaps. If any step fails, the old agent stays in use.
Create servers from a cloud account
No server yet, or need many at once? Connect a DigitalOcean, Vultr, or IDCloudHost account and servers are created straight from Saka with the agent installed automatically. No install command to copy.
- Create an API token in your provider's dashboard (DigitalOcean: API, Generate New Token with write access; Vultr: Account, API; IDCloudHost: Account, API Access).
- In the panel, open Settings, the Cloud accounts card, and paste the token. It is checked with the provider first and stored encrypted.
- Open Servers, press + Add server, choose the Create in a cloud account tab, pick the account, location, and size, then press Create server. The server shows up once its agent connects (usually 1-3 minutes).
Your provider bills the server to your account from creation until you delete it in the provider's dashboard. Removing a cloud account from Saka does not delete servers already created; also revoke the token in the provider's dashboard if you no longer use it. Your AI agent can do step 3 over MCP (server_buat), but tokens are only added in the panel, never over chat.
Snapshot, resize, and delete from the panel
On the server page, the In your cloud account card appears for servers in a connected cloud account, including servers not created through Saka (matched by public IP; on IDCloudHost by hostname).
- Snapshot now: saves the whole disk while the server keeps running. Snapshots are listed on the card and can be deleted. The provider bills their storage.
- Resize: pick mini, small, medium, or large. The server is briefly off (±1-3 minutes); Vultr can only upgrade. Take a snapshot first is on by default.
- Delete server at the provider: the VM is deleted at DigitalOcean, Vultr, or IDCloudHost and removed from Saka, so billing stops. You type the server name to confirm. On IDCloudHost its public IP is released too so it is not billed on. Servers still in a database cluster or load balancer are refused.
Supported OSes
Every OS below was fully tested on 28 Sep 2026 on a 1 vCPU / 2 GB VPS: install the agent, install OpenClaw, default security, create swap, turn on the firewall, check from outside (22 open, app port closed), then Lepas server (Release server) until clean.
| Family | Versions | Notes |
|---|---|---|
| Ubuntu | 22.04, 24.04, 26.04 | Firewall uses ufw. |
| Debian | 13 | Debian 12 could not be tested yet, so it is not on the list. |
| AlmaLinux | 8, 9, 10 | Firewall uses firewalld, Docker from the official Docker repository. |
| Rocky Linux | 8, 9, 10 | Same as AlmaLinux. |
| CloudLinux | 8, 9, 10 | Same as AlmaLinux (RHEL family). Not tested end to end yet. |
| CentOS Stream | 9, 10 | Same as AlmaLinux. |
Windows (WSL): the agent can be installed on Ubuntu 22.04 or newer in WSL 2. An older distro (e.g. Ubuntu 20.04) doesn't need upgrading: install Ubuntu 24.04 alongside it from PowerShell with wsl --install -d Ubuntu-24.04, then run the install command there. systemd must be on (the default for Ubuntu 24.04 on WSL; for other distros put [boot] systemd=true in /etc/wsl.conf, then wsl --shutdown). WSL stops when Windows shuts down or sleeps, so it suits trying things out and office use, not a 24-hour server.
3. The Readiness card
The server detail page has a Kesiapan server (Server readiness) card. The agent checks the server every 5 minutes and explains what is missing, along with how to fix it:
| Check | What it looks at |
|---|---|
| Operating system | Whether the OS is one that has been tested. |
| RAM | Enough for the agent only, for Hermes (1 GB), or for OpenClaw (2 GB). |
| Disk | Free space. A warning if it is below what OpenClaw needs, an error if the disk is almost full. |
| Swap | Present or not. Can be fixed in one click. |
| Firewall | Active or not. Can be fixed in one click, with a port check first. |
| Server clock | Synced with NTP or not (important for SSL certificates and logs). |
| Updates | Whether there are system updates waiting for a reboot. |
| Docker | Installed or not. Installed automatically when you install your first app. |
The card does not make you scroll a long list: problems come first, checks that are already fine are folded behind a button, and the fix steps for each warning sit behind Show how. A Patch health card sums up vulnerable kernel, PHP and Node.js versions and pending reboots, one line per server. A firewall owned by another hosting panel (iptables rules with a DROP policy) is recognised as an active firewall, so Saka gives no false warning and does not install a second firewall on top of it.
Web ports: servers that already run CasaOS, nginx, or Apache
A VPS that previously ran CasaOS, nginx, or Apache uses port 80, so sites from Saka cannot use a regular web address with automatic HTTPS. The Web ports (80/443) row in the Readiness card shows that and offers a Set up button. One button, no technical choices:
- Saka shows its plan first. You press Set up now.
- The old service's settings are backed up, then it is moved to another port (e.g. 8081) and restarted for a few seconds.
- Saka takes ports 80/443 and forwards every address that is not a Saka site to the old service. The CasaOS dashboard or old site stays reachable from the server address as usual, and on its new port too.
- Create sites from Saka as usual; HTTPS is set up automatically.
If any step fails, everything is restored automatically to how it was. When you release the server, the old service is put back on port 80. No data is deleted.
- What can be moved automatically: CasaOS, nginx, and Apache that only use port 80. If the old service also uses port 443 (HTTPS), a Docker container publishing port 80 (e.g. Nginx Proxy Manager, Coolify), or another program, Saka does not touch it and tells you why: use another clean server, or stop that program first.
- From MCP:
server_perbaikiwith kindweb-cek(the plan, read only) thenweb(with your approval). - In real tests (6 Oct 2026, Ubuntu 24.04): CasaOS, nginx, and Apache were moved, served through Saka with the old Host headers still arriving, and restored; a simulated failure restored everything; a site from Saka got a Let's Encrypt certificate while the CasaOS dashboard stayed reachable by IP address.
Fix swap
Press Perbaiki (Fix) on the Swap row. Saka creates a 2 GB swap file (1 GB if disk space is tight) that stays active after a reboot, with low swappiness so swap only acts as a buffer. Running apps are not interrupted. Swap matters on small VMs so apps are not force-killed when RAM runs out.
Fix the firewall: ports are checked first
- Press Perbaiki (Fix) on the Firewall row. Saka first shows the ports currently in use on the server's public address, along with the programs using them.
- Tick the ports that must stay reachable from the internet. SSH is always left open so you do not lock yourself out.
- Press Aktifkan firewall (Turn on firewall). All other ports are closed.
Ubuntu and Debian use ufw, the RHEL family uses firewalld. After the firewall is on, the agent makes sure app containers can still reach the internet (for example Telegram or your AI provider). If they cannot, the firewall is turned off again.
A firewall turned on by Saka is recorded as created by Saka, so it is turned off again when you release the server. A firewall that existed before Saka is not touched.
Bandwidth usage
The Bandwidth card on the server page answers "how much bandwidth did I use this month": data in (received by the server) and out (sent by the server) for today, this month, and last month. Show details adds a daily chart (62 days), monthly totals (13 months), and usage per site this month, so you can see which site uses the most.
- Counted by the Saka agent from the server's public network interface every minute (for servers behind NAT: the default route interface). Internal Docker, VPN, and private network traffic is not included. Reboots do not lose the count.
- Per site comes from the web access log: response size sent to visitors and uploads received. System downloads, backups, and other apps only show in the server total.
- Figures start when the Saka agent starts counting (the date is shown on the card). Dates use the server's time zone. Your provider may count slightly differently.
Quota & alerts. Use Set quota with your provider's plan (GB per month, in + out). Saka lets you know over Telegram (or email when Telegram is not linked) when this month's usage reaches 80% and 100%, once each per month. Even without a quota, Saka reports a daily spike (at least 5 GB and five times this month's daily average), a common sign of a site being scraped by bots or a compromised server sending spam.
API: GET /api/v1/server/<id>/bandwidth, and the quota with PATCH /api/v1/server/<id> {"kuota_bw_gb": 2000}; this month's summary is also in the server status. AI agents use the MCP tools server_bandwidth and server_kuota_bandwidth.
Projects: a whole app in one view
The Projects menu groups the sites, Git apps, databases, load balancers, and tunnels that make up one app, e.g. an online shop's API + web + database, even when they are spread over several servers. A project shows every service with its type (web app, worker, cron, static site, cluster database, app database), domains, state, and CPU/RAM use against its server's capacity. Workers, cron jobs, and app databases of a Git app appear automatically. A project only groups things: removing a service or deleting the project never deletes or stops anything. A project can also hold Kubernetes clusters and servers. It opens with a one-line summary (good, almost good, or problems) and what needs attention; a database cluster shows each node (role, state, lag) with its readiness, and a Kubernetes cluster shows each node (pods, CPU, RAM), problem pods, and Saka's advice. AI agents can use proyek_daftar, proyek_status, and proyek_atur.
4. Telegram notifications
Until Telegram is linked, important alerts go to your email, and each of those emails has a Link Telegram button. The panel also shows a suggestion at the top (close it with Later) with how many alerts have gone to your email. To link it:
- Open Pengaturan (Settings) in the panel, then press Hubungkan Telegram (Connect Telegram).
- Telegram opens in a new tab. Press Start on the Saka bot. The link is valid for 30 minutes.
- The settings page changes to Tersambung (Connected) by itself.
Telegram alerts come with buttons to handle them from your phone: a stopped app has View log (last 40 lines, tokens and passwords masked) and Restart (asks for confirmation first and is recorded in the Action history), a site or address that does not open has Check again now (from two locations), and every server alert has Open in panel. Buttons only work from the Telegram linked to your account and last 7 days. Send /uji to the bot to see an example.
What gets sent to your Telegram:
- A server has stopped contacting Saka Panel (no heartbeat for more than 3 minutes; not counted while the agent is updating itself), and a message when it reconnects. When 3 or more servers disconnect or reconnect together, you get one combined message instead of one per server.
- Disk usage at 90% or more: Saka immediately removes what is safe to remove (unused Docker images, build and package caches, old system logs; Git app rollback images and data are not touched), then tells you how much space was freed. Another message follows when it drops back below 85%.
- An app has stopped or is unhealthy, and a message when it runs again.
- Site error spikes, failing cron jobs, scheduled jobs that stop reporting, sites/tunnels that cannot be opened, WordPress security findings, failed backups, and HTTPS certificate problems.
- Results of long tasks (for example app installs), database clusters and load balancers that are ready or failed, and database primary switchovers.
No Telegram linked? Every important alert above still goes to your account email (once verified). Light notices such as "agent updated" are Telegram only.
Each problem is sent once, followed by one message when it recovers along with how long it lasted, so it does not become spam. The Saka bot currently only sends alerts. Commands and chatting with AI through the Saka bot will come later.
Next
Open your server and choose + Pasang aplikasi (Install app) for OpenClaw, or add a few more servers for multi-location databases and a load balancer. Have a site on cPanel or an old VPS? See Moving from cPanel and Move to another server.
Need a feature that does not exist yet? Suggest it from the Request a feature menu in the panel. Moving from AWS, Google Cloud, Azure, or other hosting and want help? Fill in the Migration help form. Details are in Feature requests & migration help.