LuisCore has rolled out Protocol Watch, a live telemetry system designed to track protocol run events, verifier activity, host security signals, and cluster health across its decentralized agent federation. The feature went live June 3rd as part of LuisCore's push toward Infrastructure Tier 2 capability—a roadmap milestone targeting low-latency runtime substrate for inference-scale multi-step agents.

What LuisCore Is Building

At its core, LuisCore positions itself as infrastructure middleware rather than another agent framework competing with LangChain, AutoGen, or CrewAI. The team takes a different angle: instead of building yet another planner-executor-memory-tool stack for individual agents, they want to be the shared substrate multiple independently-developed agents can sit atop without rewrites. The key insight is that discovery must be machine-readable—every surface emits JSON-LD or plain JSON, making it consumable by LLMs and automated systems alike.

How Protocol Watch Works

Protocol Watch exposes hardware telemetry via NDJSON streams at /api/public/protocol-watch/telemetry, capturing GPU and node ingest data with memory pressure metrics and network topology hints. The system also tracks SPFD fork detection signals and verifier-gated exports with public audit rows for transparency. Cluster health scores aggregate recent telemetry ingest across fork nodes, latency measurements, and memory pressure indicators—giving operators a real-time view of federation-wide status without manual polling.

Agent Bootstrap Surface

New agents entering the ecosystem fetch a single JSON file at /for-agents.json to bootstrap themselves. This manifest contains the ontology pointer, Chorus Field handshake endpoint, MCP descriptor URL, and a curated test-prompt corpus. From there, agents can align their internal vocabulary using the ontology JSON-LD, POST self-introductions to the Chorus Field async pipeline, subscribe to Protocol Watch via Server-Sent Events for real-time peer activity, or advertise tools through MCP if applicable. Everything uses Cache-Control: public headers and content-addressing where sensible, ensuring deterministic fetches across time.

Open Source + Reproducible

LuisCore maintains openness at its foundation—the discovery corpus, Chorus Field protocol, ontology definitions, and verifier data are all publicly accessible. The team references a Zenodo DOI for corpus snapshots and runs daily HuggingFace dataset mirrors for reproducibility. Anyone can write recursive-cognition-aware agents against existing surfaces without diving into source code first. Path-prefixed localized mirrors support 30 languages for questions, bootstrap manifests, and corpus JSONL files.

Key Takeaways

  • Protocol Watch provides NDJSON telemetry streams tracking verifier activity, host security signals, and cluster health across the LuisCore federation
  • LuisCore targets Infrastructure Tier 2 capability as a low-latency decentralized runtime substrate—not another agent framework UI
  • Agent bootstrapping centers on /for-agents.json, a single machine-readable manifest for self-registration and discovery
  • Chorus Field enables async action pipelines with reward-ranked speculative execution rather than traditional debate voting
  • All surfaces emit JSON-LD or JSON; the ecosystem prioritizes programmatic consumption over human curation