Signet
An identity and governance layer for AI agents where every action is wrapped in a signed, verifiable envelope and policy-gated before execution. Ships with Python and TypeScript SDKs plus MCP/A2A middleware for drop-in adoption across agent stacks.
Problem
Agents act on the world: calling tools, spending money, touching data. But nothing binds an action to a verifiable identity or a policy. Existing auth answers 'who is calling the API', not 'which agent, acting for whom, is allowed to do this specific thing'.
Approach
- Wrap every agent action in a signed, verifiable envelope before it leaves the process.
- Gate execution on policy: actions are checked against rules before they run, not audited after.
- Shipped Python and TypeScript SDKs so enforcement works in either stack.
- Built MCP/A2A middleware so existing agent stacks adopt it without rewiring orchestration.
Outcome
A governance layer where every action is attributable to a signed identity and policy-gated pre-execution, adoptable across agent stacks without rewriting orchestration.
Architecture
const action = await signet.sign({
agent: "researcher-7",
actFor: user.id,
tool: "payments.transfer",
args,
});
await signet.enforce(action, policies);
// verify signature + policy BEFORE run
await tools.execute(action);Highlights
Wraps every agent action in a signed, verifiable envelope.
Policy-gated execution: actions are checked before they run.
Python and TypeScript SDKs for cross-stack adoption.
MCP/A2A middleware for drop-in integration across agent stacks.
Stack
- Python
- TypeScript
- FastAPI
- Next.js
- Model Context Protocol