Open source · Apache-2.0 · npm i @11ai/execution-governance

Authorize every agent action before it runs.

Execution Governance™ evaluates each AI agent tool call against your policy and returns allow or deny before execution. Fail-closed. A signed receipt on every decision.

Request -> Verify -> Allow or Deny -> Execute -> Proof

npm install @11ai/execution-governance

Read the code Verify a live decision

The 30-second proof

A prompt injection, denied before it runs.

execution-governance demo

$ npm install @11ai/execution-governance
added 2 packages, 0 vulnerabilities
$ node demo.mjs
DENY   exfiltration: outbound call carrying secret material
       receipt 019f78c7-...  signed: true
       the http.post never ran. Nothing left the machine.
ALLOW  safe GET allowed

The agent obeyed the injection. The gate denied it. Every decision is a signed receipt you can verify yourself. Real output from a clean install.

Pre-execution authorization

The decision happens before the action, not after. A denied call never runs.

Fail-closed

Any error, timeout, unreachable engine, or missing decision is a deny. There is no fail-open path.

Cryptographic proof

Every decision is canonical JSON, sha3-512 hashed, Ed25519 signed, and chained.

MCP gateway

Gate an existing MCP server by changing one line of config.

mcp config

{
  "command": "npx",
  "args": ["-y", "@11ai/mcp-gate", "--policy", "eg-policy.yaml", "--", "node", "their-server.js"]
}

The proxy passes initialize, tools/list, and notifications through untouched. A tools/call is gated. A denied call is answered with a JSON-RPC error and never reaches the server.

Measured, not claimed