AI agents
Agents, without overclaiming.
Six roles are planned. This page shows what exists in source today. No agent controls keys, funds or consensus, and no internal endpoint is exposed.
AI agents: overview
| Role | Status | What exists |
|---|---|---|
| Economic Agent | NOT IMPLEMENTED | No source exists. Economic policy is set by owner decisions and contracts. |
| Compute Scheduler | LOCALLY VERIFIED (NATIVE-AI-ROUTING-POLICY-LOCAL) | Customer routing and registry client; a scheduler agent is not implemented. |
| Security Agent | NOT IMPLEMENTED | No source exists. |
| Training Coordinator | LOCALLY VERIFIED scoped (NATIVE-TRAINING-COORDINATION); G7 deferred | Native training coordination and dataset manifests (scoped); G7 deferred. |
| Network Monitor | Deployed on testnet | QL-007 on seed1 since 18 November 2026: a small local model (Qwen3-0.6B, CPU) checks node health every 2 minutes, writes alerts and logs operation proposals for a human (act mode since 19 November). It holds no keys or funds and sends no transactions; restarts and resyncs stay proposals. Prometheus alerts (CMP-MONITORING) run alongside. |
| Developer Agent | LOCALLY VERIFIED (ISOLATED-DEVELOPMENT-PROPOSAL-EXECUTOR) | Isolated development-proposal executor. |
Health states for agents are not published: no agent exposes a public health endpoint.