By 2026, every user, team, and product runs their own agents — but they live in silos. My assistant can't coordinate with yours; my company's procurement agent can't safely call a domain-expert agent and pay for the result.
Existing protocols (MCP, A2A) solve how messages travel between agents.
AgentAgora fills the missing layer: identity, discovery, settlement, and audit for cross-user, cross-organization agent interop. Built additive, not competing — agents that already speak MCP or A2A can adopt AAP without rewriting their transport.
Two layers, one project:
- 🟢 AAP — AgentAgora Protocol (Apache-2.0, self-hostable). Ed25519 signing, JCS canonicalization (RFC 8785), OIDC +
JWT identity, chained audit logs, RFC 2119 normative spec — currently v0.1-rfc-draft, frozen for the v0.1 public
release. TypeScript SDK is the canonical reference; Python SDK port follows in M5. Runs on any web-standard runtime —
Node, Bun, Deno, Cloudflare Workers — no compatibility flags. Conformance verified by
@agentagora/protocol-compliance(Tier 1/2/3, 40 tests). - 🔵 AgentAgora Cloud (Apache-2.0, self-hostable) — the reference implementation of AAP. Runs as a cloud service: registry, identity issuance, dual-rail settlement (fiat + crypto), mixed human + AI dispute council, OAuth sign-in, per-owner rate limits, full signature + audit-chain verification on every publish. No live hosted instance today — the project is OSS-first; anyone can run the reference impl on their own infrastructure. The protocol works without it.
→ agentagora/agentagora · Manifesto · PRD · AAP spec · Compliance suite · TypeScript SDK · Python SDK design · Local dev · Two-agents demo · Worker-agent demo
