Deploy OpenClaw
Self-hosted OpenClaw gateway — a personal AI assistant / agent runtime you connect your own LLM key and chat channels to.
Deployed 1 time on Lizard
What's included
How to self-host OpenClaw elsewhere
This is what deploying OpenClaw by hand looks like — useful context for what clicking "Deploy Now" above does automatically.
- 1Get the source. Clone
https://github.com/openclaw/openclaw(branchmain). - 2Set environment variables. Configure
OPENAI_API_KEY,ANTHROPIC_API_KEYto point at that database. - 3Start it and expose the port. Run the container and route traffic to port
18789.
On Lizard, all of the above happens when you click Deploy — the database is provisioned and its credentials are wired into the app's environment automatically.
How OpenClaw compares to other automation platforms
OpenClaw vs. Open WebUI
- Scope:
- OpenClaw is a standing gateway with channel integrations (Telegram/Discord/Slack) and agent tooling; Open WebUI is primarily a chat interface in front of a model backend.
- Always-on:
- OpenClaw runs as a persistent process that can message you first; Open WebUI is a UI you open when you want to chat.
OpenClaw vs. LibreChat
- Focus:
- LibreChat centers on a multi-provider chat UI; OpenClaw centers on the gateway/agent layer, with channels and tool access built around it.
- Deployment shape:
- Both are self-hostable and open source — the difference is mainly what you're deploying it to do.
OpenClaw vs. AutoGPT
- Maturity model:
- OpenClaw is built around a stable gateway process with defined auth and channel plugins; AutoGPT-style frameworks lean more experimental/autonomous-loop oriented.
- Day-to-day use:
- OpenClaw is designed to be something you talk to daily through a channel you already use; AutoGPT-style tools are typically run for a task, then stopped.
Ready to deploy OpenClaw?
Deploy NowWhy deploy OpenClaw on Lizard
OpenClaw is a self-hosted, MIT-licensed personal AI assistant and agent gateway — you connect your own LLM key (Anthropic, OpenAI, Gemini, or a local model) and, optionally, chat channels like Telegram, Discord, or Slack, and it runs as a long-lived process rather than a per-request API call. Because it's open source, self-hosting doesn't trade away features for cost the way it can with a closed SaaS tool — you're running the exact same gateway, just on infrastructure you control. This template handles the part that trips up most Docker deployments: OpenClaw's gateway binds to loopback and refuses to start exposed without authentication by design, so a naive `docker run -p` gets you an unreachable container. This template's startup script generates a random gateway token on boot and prints it to the deploy logs, so it comes up reachable and authenticated on the first try.
Lizard vs. other VPS providers for OpenClaw
| Provider | With Lizard | With the other provider |
|---|---|---|
| Linode | This template already builds OpenClaw's real multi-stage Dockerfile, exposes the gateway on a non-loopback bind, and generates your auth token automatically — deploy and read the token from the logs. | A bare VPS: install Docker, clone the repo, build the multi-stage image yourself (it needs real memory headroom to compile), then work out that the gateway won't accept external connections until you explicitly override its bind address and set an auth token. |
| Vultr | The gateway is reachable at a generated HTTPS domain within minutes of clicking Deploy, with TLS already handled. | Inexpensive compute, but you own TLS termination, the Docker build, and keeping the process alive after a crash or reboot. |
| AWS Lightsail | One click produces a running, authenticated OpenClaw gateway — no IAM, security groups, or manual port-forwarding config. | Full AWS ecosystem access, at the cost of meaningfully more networking and access-control setup for a single gateway process. |
Common use cases
- A personal AI assistant reachable from Telegram, Discord, or Slack, backed by your own LLM key instead of a per-seat SaaS subscription
- A standing agent gateway for coding/dev-tool workflows — connect it to your own tools and let it operate over a persistent session rather than one-off API calls
- A private, self-hosted alternative to a hosted AI assistant product, where the model, memory, and connected tools all stay under your control
- A base to build on top of — OpenClaw ships as a plugin-extensible gateway, not a fixed feature set
Cost of self-hosting OpenClaw on Lizard
Self-hosting via this template costs whatever compute and memory it actually uses, billed by the second at Lizard's published rate — not a fixed subscription:
| Usage | Profile | Est. monthly cost |
|---|---|---|
| Personal / light use | idle most of the day, occasional chat/agent sessions (~0.1 vCPU / 400MB average) | ~$6.00/mo |
| Active daily assistant | multiple channels connected, frequent sessions (~0.3 vCPU / 800MB average) | ~$14.01/mo |
Estimated from Lizard's published per-second compute rate at a realistic average utilization, not a maxed-out allocation — actual cost tracks your real workflow volume. See /compare-cloud-prices for the full methodology.
Ready to deploy OpenClaw?
Deploy NowFrequently asked questions
Is this gateway exposed to the public internet?
Yes — like Railway's own OpenClaw template, this template gives the gateway a public HTTPS domain so channels and your browser can reach it. It is not open by default: a random auth token is generated on first boot and printed to the deploy logs (Dashboard → Logs), and the gateway refuses to start without it. Treat that token like a password — anyone with it can control the gateway.
Where do I find my gateway auth token?
Check the app's deploy/runtime logs right after it starts — the startup script prints it once as "OpenClaw gateway auth token (save this — required to connect)". It is not stored anywhere else, and this template has no persistent volume, so a new token is generated on every restart.
Do I need to add my own AI API key?
Yes. OpenClaw itself is free and open source, but it calls out to an LLM provider you control — set ANTHROPIC_API_KEY, OPENAI_API_KEY, or another supported provider key as a variable on this template. Without one, the gateway still runs, it just can't answer.
Does state (chat history, connected channels) survive a restart?
Not on this template yet — Lizard's template system doesn't currently support attaching a persistent volume to an app the way Railway's OpenClaw template does with its required /data volume. A restart resets local state and rotates the auth token. If you need persistence, run OpenClaw somewhere with a mounted volume, or ask about volume support for this template.
How much does self-hosting OpenClaw on Lizard cost?
There's no flat fee — Lizard bills by the second for the CPU and RAM the gateway actually uses. For light personal use that's typically a few dollars a month; the bigger cost driver is usually your LLM provider's token usage, not hosting. See the estimate above for the hosting-only portion.