Autonym

Self-certifying identity and messaging for AI agents.

Overview

Autonym is an open protocol for AI agent identity that is self-certifying, portable, and federated. Agents derive their identifier from their own Ed25519 key material — no registry, domain, or blockchain required. A key event log (KEL) with mandatory pre-rotation provides post-compromise recovery: even if a key is stolen, the attacker cannot hijack the identity. Autonym nodes federate to replicate KELs and provide store-and-forward messaging with end-to-end encryption (X25519 + XChaCha20-Poly1305). All protocol data uses deterministic CBOR encoding (RFC 8949).

Repository structure

AUTONYM.md          Protocol specification (living document)
paper/
  main.tex          Research paper entry point
  sections/         Paper body (introduction, protocol, security analysis, etc.)
  appendices/       CBOR schemas, error codes, test vectors
  scripts/          Test vector generator
  autonym.bib       Bibliography
  shared-macros.tex Shared notation macros

Research paper

The paper/ directory contains a formal research paper covering the protocol design, security analysis, and performance evaluation.

Build with:

cd paper
pdflatex main && biber main && pdflatex main && pdflatex main

Output: paper/main.pdf

Test vectors

Deterministic test vectors for inception, rotation, deactivation, and end-to-end encryption.

pip install pynacl cbor2 blake3
python paper/scripts/generate_vectors.py

License

TBD