Skip to content

πŸ™Œ OpenHands: AI-Driven Development

Update SELF_HOSTING.md

Robert Brennan0740d20c8728
diff --git a/SELF_HOSTING.md b/SELF_HOSTING.md
index 967e74c809..84daadeef3 100644
--- a/SELF_HOSTING.md
+++ b/SELF_HOSTING.md
@@ -144,68 +144,6 @@ The ingress proxy binds to `127.0.0.1:8000`, and the underlying agent server,
automation backend, and Vite dev server all bind to `127.0.0.1` as well β€” so
they're only reachable from the machine itself.
-### `SESSION_API_KEY` is auto-generated
-
-On first run, the dev scripts generate a random `SESSION_API_KEY` and persist
-it to `~/.openhands/agent-canvas/session-api-key.txt`. The same value is wired
-into both the agent server (which refuses any `/api/*` request without a
-matching `X-Session-API-Key` header) and the frontend (which sends it on every
-request), and it stays stable across restarts. You don't need to configure
-anything.
-
-To rotate the key, delete the file and restart the app. To pin a specific
-value instead of the auto-generated one, export `SESSION_API_KEY` before
-starting β€” it takes precedence over the persisted file.
-
-> [!IMPORTANT]
-> Treat the key file like an SSH private key. Anyone who learns its value can
-> drive your agent. Keep it on the machine and don't copy it elsewhere.
-
-### Access the UI over an SSH tunnel
-
-With just steps 1–3 done, the simplest way to use the UI is to forward the
-ingress port over SSH from your laptop:
-
-```bash
-ssh -L 8000:127.0.0.1:8000 your-vm
-# then open http://localhost:8000 in your browser
-```
-
-No public ports, no nginx, no TLS to manage β€” the SSH connection is what
-authenticates and encrypts you. This is a great starting point and a good
-fallback for debugging.
-
-### Keep it running across reboots
-
-Wrap the command in a `systemd` unit, a `tmux`/`screen` session, or a process
-manager like `pm2`. A minimal Linux `systemd` unit:
-
-```ini
-# /etc/systemd/system/agent-canvas.service
-[Unit]
-Description=Agent Canvas
-After=network.target
-
-[Service]
-Type=simple
-WorkingDirectory=/root/agent-canvas
-ExecStart=/usr/bin/npm run dev:dangerously-dockerless
-Restart=on-failure
-User=root
-
-[Install]
-WantedBy=multi-user.target
-```
-
-```bash
-systemctl daemon-reload
-systemctl enable --now agent-canvas
-journalctl -u agent-canvas -f
-```
-
-On macOS, use `launchd` (a `LaunchAgent`/`LaunchDaemon` plist) for the same
-effect.
-
## 4. (Optional) Get a domain and put nginx + Let's Encrypt + basic auth in front
If you want to reach the UI from a browser without an SSH tunnel β€” for
@@ -377,58 +315,3 @@ open against the remote domain.
> cross-origin `/api/*` calls made by your local Agent Canvas. Alternatively,
> skip basic auth and rely on the cloud firewall + `SESSION_API_KEY` for
> protection.
-
-## Operational tips
-
-- **Rotate the basic-auth password and `SESSION_API_KEY` periodically**,
- especially after offboarding anyone who had access. `htpasswd` changes take
- effect on the next request β€” no nginx reload needed. To rotate the session
- key, delete `~/.openhands/agent-canvas/session-api-key.txt` and restart the
- app.
-- **Keep the VM patched** (`unattended-upgrades` is fine for Ubuntu).
-- **Back up `~/.openhands-dev/` (or wherever conversations are persisted)**
- if you care about conversation history. Treat the backup as sensitive β€” it
- contains transcripts, secrets, and possibly auth tokens.
-- **Watch the logs.** `journalctl -u agent-canvas`,
- `/var/log/nginx/access.log`, and `/var/log/nginx/error.log` are your friends
- when something looks off. Repeated 401s in the access log are normal scanner
- noise; a successful 200 from an IP you don't recognize is not.
-- **Do not expose the ingress (`:8000`) or backend ports (`:18000`, `:18001`,
- `:3001`) directly.** Only `:443` (via nginx) should be reachable from
- outside. If you need to debug, tunnel over SSH
- (`ssh -L 8000:127.0.0.1:8000 vm`) rather than opening a port in the firewall.
-
-## Advanced: defense in depth
-
-The cloud firewall (step 2) is your primary network perimeter. If you want an
-extra backstop in case it is ever misconfigured β€” for example, a teammate
-relaxes a rule, a Terraform diff lands wrong, or you migrate to a provider
-whose firewall fails open during maintenance β€” also enable a host-level
-firewall on the VM itself.
-
-With `ufw` on Ubuntu:
-
-```bash
-ufw default deny incoming
-ufw default allow outgoing
-ufw allow 22/tcp
-ufw allow 80/tcp
-ufw allow 443/tcp
-ufw enable
-```
-
-And confirm the ingress, agent server, automation backend, and Vite ports are
-bound to `127.0.0.1` only, so even if a firewall rule slips, none of them are
-reachable from outside the host:
-
-```bash
-ss -tlnp
-# expect 127.0.0.1:8000 (ingress)
-# 127.0.0.1:18000 (agent server)
-# 127.0.0.1:18001 (automation backend)
-# 127.0.0.1:3001 (Vite dev server)
-# never 0.0.0.0:* or :::* for any of those
-```
-
-If you ever see one of those ports listening on `0.0.0.0` or `::`, stop the
-app immediately and investigate before re-enabling it.