A budget your AI agent can't overspend. A record it can't fake.

Leash gives every AI agent a signed pass with a hard money limit, keeps your API key encrypted where the agent can't reach it, and seals every call into a record nobody can quietly rewrite.

Watch a real agent hit its limit Self-host from GitHub

The demo uses a real AI and real money, in fractions of a cent, paid by us. No sign-up needed.

Your agent holds only a Leash pass lsh_eyJhZ2VudCI6…
Leash your real key stays here
  • Genuine pass, not expired
  • Within its money and call limits
  • Not stopped
  • No trap keys, no leaked secrets
  • Written to the record
Your AI provider sees only approved calls

Why we built Leash

$100 budget. $5,000 bill. Logs that said everything was fine.

A developer left an AI agent to finish a task with a $100 budget. When he came back, it had spent about $5,000, doing far more than he asked. When he checked its logs, they said nothing had gone wrong.

  1. The budget was a request, not a limit. Leash enforces it.

    The agent gets a pass signed with HMAC-SHA256 that carries its budget. Before each call, Leash works out the most that call could cost and refuses it if it could go over. The agent can't spend past its limit, and it can't edit the pass to raise it.

  2. The agent held the real API key. Leash keeps it out of reach.

    Your key is encrypted with AES-256-GCM and only unlocked inside Leash for the moment of each approved call. The agent only ever sees its pass.

  3. The agent wrote its own logs. Leash writes the record.

    Every call is recorded by Leash, not the agent, and each entry is sealed to the one before it with HMAC-SHA256. Change one entry and Leash points to exactly which one.

See it with a real agent and real money Give it a task and a budget of a tenth of a cent. It's told to keep improving its work, so it overdoes it. Watch Leash stop it at the limit, then try to fake the record.

Put your agents on a leash

Watch a real agent hit its limit and try to fake the record in the live demo, then run Leash for your own agents from the source code.

Watch a real agent hit its limit Self-host from GitHub