Open infrastructure for agents that act.
Agent infrastructure is the reusable technical substrate that enables AI agents to act safely, reliably, and continuously in real-world environments.
It provides the execution environments, communication, state, data access, authority, observability, recovery, evaluation, and human control that surround the model. It is not a particular agent application or framework; it is the shared foundation on which many agents can be built and operated.
| Area | What it includes |
|---|---|
| 🧪 Execution Environments & Sandboxing | Sandboxes, browsers, containers, filesystems, process isolation, and resource controls |
| 🤝 Agent Communication & Coordination | A2A protocols, discovery, messaging, delegation, shared context, task routing, and conflict resolution |
| 🧠 Memory, Data & Filesystems | Memory, metadata, artifacts, retrieval, shared state, and agent-native data primitives |
| 🔑 Identity, Permissions & Trusted Tools | Agent identity, delegated authority, credentials, capability-based access, and safe tool interfaces |
| 🔄 Durable Workflows & Recovery | State, checkpoints, retries, scheduling, persistence, and recovery |
| 🔎 Observability, Auditability & Provenance | Logs, traces, replay, action history, provenance, and policy decisions |
| 🙌 Human Handoffs & Control | Approvals, escalation, intervention, context transfer, and accountability |
| ✅ Evaluation & Operational Reliability | Testing, simulation, containment, recovery, consistency, and human-control evaluation |
We welcome projects, protocols, prototypes, and ideas that make agent systems more reliable, legible, interoperable, and accountable.
