How Much Does It Cost to Self-Host an AI Agent?
For a single person, expect under $100 a month: a cheap VPS from roughly $5 to $20, plus LLM API usage that varies with how hard your agent works. The software itself (OpenClaw) is free and MIT-licensed; the real cost is the model, not the hosting.
The headline number for self-hosting a personal AI agent is comfortably under $100 a month, and often far less. The software is free: OpenClaw is MIT-licensed and open source. What you pay for is the server it runs on and the LLM doing the thinking.
The catch is that the LLM bill is variable, not fixed. A quiet agent that sends a morning briefing costs cents a day; a busy one running frequent heartbeats and long reasoning tasks costs more. So budget a range, not a single figure.
How it works
Three line items make up the bill, plus one that is easy to forget.
- Hosting. A VPS that meets OpenClaw’s minimum specs (2 cores, 2 GB RAM, 2 GB storage) runs from roughly $5 to $20 a month. A managed one-click platform like Railway sits at the higher end of that but removes server admin.
- LLM API. This is the swing factor. You pay per token to your chosen provider (Claude, GPT, Gemini or DeepSeek). DeepSeek is the cheap option; frontier Claude or GPT models cost more but reason better. Frequent heartbeats and long tasks raise the bill.
- Channels and integrations. Telegram is free. WhatsApp Business and a dedicated number may carry small fees depending on how you connect.
- Your time. Not a cash cost, but real: plan for 1 to 3 hours a month of maintenance. At any sensible hourly rate, that is the largest hidden cost.
Worked example
A realistic monthly budget for one person, compared with a managed service:
| Item | Self-host OpenClaw | Liv (managed) |
|---|---|---|
| Software licence | Free (MIT) | Included |
| Hosting | $5–20 VPS | Included |
| LLM API | Variable, ~$10–60+ | Included |
| Maintenance time | 1–3 hrs/month | None |
| Security/compliance | DIY | CASA Tier 2, vaults |
| Cash total | Often under $100 | $79 (Pro) / $149 (Max) |
The honest read: self-hosting can be cheaper in cash, especially with a cheap model and a quiet agent. Whether it is cheaper overall depends on what your time is worth, since the ops hours are the part the spreadsheet hides. The self-hosted vs managed comparison works through that trade in full, and how much OpenClaw costs to run drills into the per-token side.
Try this in Liv
If a predictable bill matters more than the lowest possible floor, Liv folds hosting, LLM usage and security into one price.
- Start a 14-day free trial at app.liv4all.com, no credit card.
- Message Liv on Telegram, the default channel.
- Connect Gmail and Google Calendar via Google OAuth, revocable any time.
- Optionally link WhatsApp (invite-only, needs a dedicated eSIM).
Liv is early access with batched onboarding, so you may join a queue. Pricing is flat: Pro $79/month, Max $149/month, no per-task credits to track.
Common questions
Is OpenClaw free?
Yes, the software is MIT-licensed and free. You pay only for hosting and LLM usage. See is OpenClaw free.
What is the cheapest way to run it?
A $5 VPS plus a low-cost model like DeepSeek, with a light heartbeat schedule. Costs scale with how much the agent actually does.
Why is the LLM cost so variable?
You pay per token. Complex reasoning, long context and frequent heartbeats all consume more tokens, so a busy agent costs noticeably more than a quiet one.
How does this compare to credit-based tools?
Credit-priced agents like Manus can burn 500 to 900 credits on a single complex task, making costs hard to predict. A self-hosted setup at least bills transparently per token.
Does a managed host save money?
Cheap managed OpenClaw hosts (around $12–22/month) save you server admin but still leave the LLM bill and add little security. A fully managed service trades a higher flat fee for zero ops.
What is the real total cost of ownership?
Cash plus time. Add your maintenance hours to the cash figure before comparing against a managed flat fee.