Open infrastructure for the Agent Settlement layer.
| Payments | How value moves. |
| Authorization | Whether an agent may spend. |
| Settlement | Whether the economic obligation was satisfied and what happens to committed value. |
A2A-SE defines settlement semantics for autonomous agent transactions: economic commitment, delivery, verification, release/refund, disputes, reputation, and finality. Zero modifications to A2A core. Currency-agnostic. The exchange is an interface, not a single vendor service.
Spec first published February 17, 2026 —
c5ba9aaa.
Docs: What is Agent Settlement? · Site: a2a-settlement.org · Run a settlement: exchange docs
| Repository | Description |
|---|---|
| a2a-settlement | A2A-SE specification, reference exchange, Python & TypeScript SDKs |
| Repository | Description |
|---|---|
| settlement-conformance | Neutral conformance suite — prove settlement-capable without our exchange |
| Repository | Description |
|---|---|
| a2a-federation-rfc | Optional cross-exchange interoperability (DIDs, VCs, peering) |
| Repository | Description |
|---|---|
| a2a-settlement-auth | OAuth 2.0 settlement scopes — spending limits, counterparty policies, delegation |
| a2a-settlement-mediator | AI-powered dispute resolution + WORM settlement pipeline |
| a2a-settlement-dashboard | Human oversight — balances, escrows, audit, revoke |
| a2a-settlement-mcp | MCP server — escrow, release, refund, reputation tools |
| settlebridge-ai | SettleBridge Gateway (product) — policy enforcement + bounty marketplace |
| mcp-trust-gateway | MCP trust evaluation above OAuth |
| otel-agent-provenance | OpenTelemetry semantic conventions for agent provenance |
| Repository | Framework |
|---|---|
| adk-a2a-settlement | Google ADK |
| langgraph-a2a-settlement | LangGraph |
| crewai-a2a-settlement | CrewAI |
| litellm-a2a-settlement | LiteLLM |
Agent A Exchange Agent B
│ │ │
├── create_escrow ────────►│ │
│ │◄── accept / work ─────────┤
│ funds held │
│ │◄── deliver ───────────────┤
├── release / refund ─────►│ │
│ ├── settle ────────────────►│
- Discover — agents find each other via A2A
- Authorize — spend permission (AP2 / OAuth scopes)
- Commit — economic commitment (escrow or equivalent)
- Execute / deliver — work + optional provenance
- Settle — release, refund, partial, or dispute
- Finality — outcome + reputation + audit trail
- Settlement layer, not a payment rail — rails move value; A2A-SE decides finality
- Zero A2A modifications — native extension
- Currency-agnostic — blockchain not required
- Conformance-first — any rail can prove settlement-capable behavior
- Federation optional — Core semantics stand alone
Historical feature notes (federation RFC, SettleBridge Gateway, OTel provenance, MCP Trust Gateway, web grounding, efficacy holdback, KYA, batch escrow, spending guards, TTL watchdog, webhooks, WORM pipeline, provenance attestation, deliver step, and framework adapters) remain available in git history and per-repo READMEs. Prefer the docs site and SPEC.md for normative material.
A project of TruthSetter LLC. Open source under MIT.