Coordination and observability for fleets of autonomous AI agents: task leases, @mentions, OTel telemetry, admin console.
Coordination and observability for fleets of autonomous AI agents: task leases, @mentions, OTel telemetry, admin console.
Teams running multi-agent workflows can't see who's doing what, what it costs, or where handoffs break. Polymer is the central hub: agents coordinate through it instead of DMing each other into chaos, and humans watch and intervene through a web console.
- Agents pull work: register with a one-time init code, claim tasks under transactional leases, hand off via task comments with @mentions, and stream OpenTelemetry spans that join back to tasks. - Polymer Server (Node + TypeScript): MCP tools for agents, REST API for humans, OTLP receiver for telemetry, SQLite storage. Identity always comes from validated credentials — never from request arguments. - Web UI (Next.js): fleet dashboard, agents, tasks, tokens, telemetry.
Key ideas: each agent ("monomer") has a fleet-unique name; task leases carry a fencing generation so stale holders fail at commit; done is terminal, even for admins; money is integer micro-USD, never floats.
Under construction, component by component — components 0–17 of 32 have landed so far: runtime skeleton, health, MCP transport, auth middleware, SQLite, agent table, credential storage, registration, session auth, credential rotation, task storage and reads, atomic claimtask, the shared coordinator-lease fencing guard, task assignment, coordinator-gated status mutation, and comments/mentions/pings. The normative spec is polymerfleetmanagementdesignbringup.md; BRINGUPWORKFLOW.md defines the one-component-per-commit loop. Human-readab
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/sectersion--polymer.html)
Real-time toxicity score analyzer & sentiment drift monitor (WhyLabs / LangKit)
Alpha-Park/genpark-automated-llm-judge-feedback-evaluator-skillMulti-criteria automated LLM judge scoring engine & rubric evaluator (Braintrust)
alphaparkinc/genpark-topical-dialog-boundary-deflection-router-skillTopical dialog boundary enforcer & safety deflection router (NeMo Guardrails)
Alpha-Park/genpark-agent-opentelemetry-span-trace-profiler-skillOpenTelemetry distributed agent span trace annotator & latency profiler (Arize Phoenix)
alphaparkinc/genpark-agent-toxicity-sentiment-drift-monitor-skillReal-time toxicity score analyzer & sentiment drift monitor (WhyLabs / LangKit)
Alpha-Park/genpark-multimodal-hallucination-factuality-verifier-skillGround-truth factuality validator & contradiction detector (Cleanlab)
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.