AI agents need more than context. They need to know who they represent, what they are allowed to do, when that authority expires, and how their actions can be verified.
Aithos is a trust layer that gives people and organizations:
- portable, cryptographic identity;
- scoped and time-bounded mandates;
- recursive delegation and revocation;
- tamper-evident, signed action logs;
- verification from files alone — a server is never the trust authority.
Status: alpha. The protocol, Rust reference implementation, TypeScript SDK and hosted services are under active development and available for testing.
- Try Aithos — create an identity and explore the hosted product.
- Read the protocol — the problem, model and normative specification.
- Explore Aithos Core — the Rust reference implementation and conformance vectors.
- Build with the SDK — the high-level TypeScript entry point for agentic applications.
| Repository | Purpose |
|---|---|
| aithos-core | Rust reference implementation: pure protocol logic, bundle storage, CLI, WASM, gateway and provider components. |
| aithos-protocol | Protocol specification, threat model, whitepaper and TypeScript reference packages. |
| aithos-sdk | High-level TypeScript SDK for identity, mandates, data, compute, wallet and assets. |
| protocol-client | Environment-agnostic TypeScript client for signing, verification and platform access. |
| aithos-app-example | End-to-end owner/delegate reference application using signed envelopes. |
A useful agent should be able to demonstrate a complete chain:
identity → authority → action → verification → revocation
Aithos makes that chain portable across models, applications and infrastructure providers.
Issues, technical criticism and interoperability work are welcome. The protocol specification and conformance vectors are public; software licenses are stated explicitly in each repository.
Website: aithos.be · Contact: mathieu@aithos.be