Verifiable coordination for AI agents.
A2AC builds infrastructure for multi-agent systems that need durable task state, signed execution receipts, human approval gates, and portable artifact integrity.
Agent systems are moving from single-chat assistants to teams of specialized models and tools. That creates a new operational problem: how do you know who did what, what changed, whether it was approved, and whether the output still verifies later?
A2AC focuses on that coordination layer:
- structured task routing
- model-specialized execution
- signed receipt ledgers
- human approval workflows
- background agent operation
- artifact integrity and provenance
- local/cloud containment boundaries
- Execution leaves evidence. Important agent actions should produce durable receipts.
- Humans approve high-risk work. Deploys, billing, publishing, and destructive changes should be explicit approval objects.
- State outlives chat. Tasks, results, approvals, and receipts should survive restarts and lost sessions.
- Agents coordinate through shared state. Work should route to the model best suited for the task.
- Artifacts should verify anywhere. Outputs should be hashable, signable, schema-bound, and portable.
- a2ac-workspace - workspace extension and durable agent coordination surface
- a2ac - public overview of A2AC coordination concepts
- cube-protocol-spec - content integrity profile for AI artifacts
A2AC is informed by ongoing work around portable agent interoperability and artifact verification.
A2AC is active infrastructure under development. Public repositories are being separated into clean protocol, extension, SDK, example, and documentation surfaces.
- Website: a2ac.ai
- Founder: Phil Hills