Works with the AI tools people already use
AgentOnRails caps what your AI agent can spend, can prove it was really your agent that paid, and keeps things like your email or card number from leaking out in a request.
They can call paid APIs, renew a subscription, book something, or check out at a store on their own. That's genuinely useful, right up until one goes over budget, pays the wrong site, or hands out an email address or card number it had no business sharing. AgentOnRails sits between your agent and the internet and stops all three.
Most people land on one of three habits. None of them actually solve it.
Safe, but now your agent can't do the useful part either: no paid APIs, no automatic renewals, no checkout.
Works for one payment. Doesn't scale to an agent making dozens of small ones a day. Now you've just re-hired yourself.
Fine, until the one time it isn't: an over-budget charge, a bad site, or a leaked detail you only find out about after.
AgentOnRails is the fourth option: a limit your agent can't cross, proof of who spent it, and a filter on what it shares. Automatic once, not manual every time.
✓Install it once. It runs quietly in the background from then on.
Decide how much your agent can spend a day, and how much on any one thing. Nothing above that ever goes through.
daily limit · per-item capWhen your agent needs to pay for something (an API call, a small tool, a subscription) it pays in seconds, without asking you each time.
paid automatically · under budgetTurn on two add-ons any time: one signs every payment so you can prove which agent made it, the other catches sensitive data before it leaves your machine.
optional · add-onsTurn on only what you need. Each one plugs into the same request your agent already makes.
Your agent's requests get paid automatically whenever a site asks for payment, never over the limit you set, per agent.
Every agent gets a signature only it has. Prove exactly which agent made a payment, and shut one off instantly if it's ever compromised.
Emails, card numbers, and API keys are caught before anything leaves your machine: block them, redact them, or just get notified.
A page on your own computer, not a cloud service, showing what got paid, who signed it, and what got blocked.
| Agent | Today | Signed by | Status |
|---|---|---|---|
| pricing-bot | $0.86 / $5.00 | did:key:z6Mk…7Fd | allowed |
| pricing-bot | - | did:key:z6Mk…7Fd | redacted · email |
| research-agent | $0.14 / $2.00 | did:key:z6Mv…2Aq | allowed |
| research-agent | - | unsigned | blocked · no identity |
Local-first by design. That doesn't change as it grows.
It lives on your own machine as a quiet background process. Nothing routes through our servers by default.
Payments move directly between your agent and whoever it's paying. We forward and sign it, we never hold it.
Your wallet and identity keys stay locked with strong encryption, on your disk.
AES-256-GCM + scrypt, never sent anywhere to unlockThe free payment engine is Apache-2.0. Every payment lands in a local log you can look at yourself.
AgentOnRails is a small proxy your agent's HTTP client already knows how to talk to. No SDK, no rewritten code. It also speaks MCP natively, so Claude Desktop, Claude Code, Cursor, or any MCP client gets the same policy-gated payment tools directly.
No account required. Four steps and your agent is running under a budget.
One command, runs quietly on your own computer. Nothing to configure yet.
brew install aorGive it a name and a wallet. AgentOnRails generates both for you.
aor agents createA daily cap and a per-item cap. Sensible defaults are already on, change them any time.
daily limit: $5 · per item: $0.10Your agent starts paying for what it needs on its own, under that limit. Watch it happen on the dashboard.
Ready when you are.
Get started freeOr build from source: go install github.com/agentOnRails/agent-on-rails/cmd/aor@latest
