Self-hosting OpenClaw needs a VPS with at least 4GB RAM (8GB if you're running browser automation), costing roughly $6-9/month. OpenClaw itself is free and MIT-licensed — the real variable cost is the LLM API it calls on every action, billed per token by whichever model provider you connect, which can exceed the hosting bill many times over.
Key takeaways
- OpenClaw the software costs nothing — it's open source under the MIT license. Every real cost here is infrastructure (the VPS) or usage (the LLM API calls it makes).
- 4GB RAM is the commonly cited practical minimum, with 8GB recommended once browser automation is in active use — a headless browser session adds real memory overhead beyond the agent process itself.
- The VPS cost is small and fixed — roughly $6-9/month for a spec that meets the minimum. It is not where the real expense lives.
- The LLM API bill is the variable, usage-scaling cost, and it's billed separately by whichever model provider (Claude, GPT, Gemini, DeepSeek, or a free local model via Ollama) you point the agent at — see the prompt-caching cost mechanics guide for how that billing actually works.
- The project has changed names twice. It was released as Clawdbot, briefly renamed Moltbot, then settled on OpenClaw — reportedly following a trademark dispute with Anthropic over the original name.
What OpenClaw actually is
OpenClaw is a free, self-hosted, general-purpose AI agent that connects to an LLM of your choice and acts on your behalf through a chat interface — typically Telegram, Discord or Slack. Depending on what you authorize it to do, it can read and act on email, manage a calendar, run terminal commands, deploy code and automate a browser. It's a thin, extensible layer that routes your instructions to a model and executes whatever the model decides to do — which is exactly why the model you choose, not the agent software, drives most of the running cost.
Two bills, not one
The mistake "how much does OpenClaw cost" makes as a question is treating it as a single number. There are two independent costs, and conflating them is how people get surprised by their bill.
| Cost | What drives it | Typical shape |
|---|---|---|
| VPS hosting | Server spec (RAM mainly), chosen once at setup | Fixed monthly, ~$6-9 |
| LLM API usage | Every agent action that calls the model, billed per token | Variable, scales with actual use |
The VPS bill is predictable because it doesn't change with how much you use the agent — a $7/month server costs $7/month whether OpenClaw is idle or handling fifty messages an hour. The LLM bill does the opposite: it's near-zero if the agent is barely used and can climb well past the hosting cost for an actively-used setup, because a capable model charged per token isn't free just because the agent wrapping it is.
The VPS side: what spec you actually need
| Use case | RAM | vCPU | Approx. price |
|---|---|---|---|
| Chat-only agent, light use | 4 GB | 1 | ~$6.49/mo |
| Browser automation in active use | 8 GB | 2 | ~$7-9/mo |
The jump from the chat-only spec to the browser-automation spec is smaller than it sounds — both land in the same rough $6-9/month band, because a budget VPS plan that doubles RAM often doubles vCPU count and storage for a modest price step, not a doubled bill. See the cheapest VPS comparison for how these specs stack up against other budget providers at the same price point.
Documented vs. inferred vs. unsupported
| Claim | Status | Basis |
|---|---|---|
| OpenClaw is open source under the MIT license | Documented | Stated in the project's own repository and consistent across independent coverage |
| The project was previously named Clawdbot, then briefly Moltbot | Documented | Reported consistently across multiple independent technology outlets covering the project's history |
| 4GB RAM minimum / 8GB recommended for browser automation | Inferred | Commonly cited figures across setup guides and hosting tutorials — not a single official hardware-requirements page confirmed directly in this pass |
| The $6-9/month VPS price range quoted here matches Hostinger's current live rate | Inferred | Sourced from third-party hosting reviews and a regional Hostinger page checked October 2026, which showed some disagreement on exact figures — confirm the current rate directly before buying |
| The LLM API bill always exceeds the VPS hosting bill | Unsupported | True for active, frequent agent use at typical model pricing, but a lightly-used agent or one pointed at a free local model via Ollama can have a near-zero LLM bill, making the VPS cost the larger of the two |
Which setup makes sense
- Trying it out, chat-only, no browser automation: the 4GB RAM minimum spec is enough — don't pay for the 8GB tier until you actually need it.
- Running browser automation regularly: budget for the 8GB tier from the start; a 4GB server under real browser-automation load is a common source of out-of-memory failures.
- Want a one-click deploy without managing Docker yourself: Hostinger's VPS application catalog includes OpenClaw as a one-click template — convenience, not a different price.
- Budgeting the whole setup, not just hosting: estimate the LLM side separately and realistically — the prompt-caching cost mechanics guide covers what actually drives a token bill up or down.