CKS ecosystem - CKS: verifiable, self-maintaining knowledge for LLMs.

by•
63 MCP tools, 4 agents, 7 sweepers. CKS gives LLMs a verifiable, self-maintaining knowledge backbone: validate, evolve, branch, merge, self-heal graphs. Detects contradictions, stale facts, outdated graphs. Agents resolve them autonomously. Offline search, plugins, SQLite/Postgres, backup, HTTP transport. 2 000+ tests. Zero API keys. Install: pip install cks-mcp

Add a comment

Replies

Best

Hey Product Hunt.

Two weeks ago, CKS was a 24-tool anti-hallucination validator. Today it's a

63-tool, multi-agent knowledge ecosystem with 2 000+ tests across five

repositories (cks-core, cks-runtime, cks-mcp, cks-studio, cks-website).

The biggest shift: CKS is no longer something you call — it's something that

runs on its own.

Seven background sweepers patrol your graphs. They find stale inference chains,

expired provenance records, facts whose valid_until has passed, outdated

registered graphs, unhealthy graphs, and logical contradictions. Every finding

goes into a shared outbox queue.

Four autonomous agents consume those tasks and resolve them. The Critic Agent

handles all conflict types — LLM arbitration for inference conflicts,

re-verification for provenance, mechanical contradiction resolution. The

Enrichment Agent searches Wikipedia and arXiv for external context and links it

back with provenance. The Fork Resolution Agent resolves CRDT forks using

causality tracking. The Pipeline Agent orchestrates Researcher → Synthesizer →

Reviewer → Arbiter steps.

We tested this on ourselves. We built a knowledge graph of the entire CKS

ecosystem from its five GitHub repos — 277 objects, 158 relations. Saved it.

Loaded it in a new conversation. Asked Claude to analyze it. It found real

bugs — self-loop dependencies, missing objects, asymmetric relations. We fixed

them through evolve_knowledge without rebuilding. The system was debugging

itself.

Also new:

• cks-studio — a React visual workspace with 2D/3D graph views, AI Chat,

agent control panels, version diff, gallery, and PWA support.

• cks-website — full docs and interactive in-browser demo, no server needed.

• Backup/restore/migrate tools: export your entire knowledge base, import

into Postgres, migrate between backends.

• Plugin framework — FastEmbedPlugin, GossipPlugin, and your own.

• HTTP transport for web frontends, plus stdio for Claude Desktop.

• CRDT replication with fork detection and quarantine validation.

• Offline semantic search via fastembed — no API keys, no network.

Install, connect to Claude Desktop (or run cks-studio), and start building

knowledge that verifies and maintains itself.

Ask me about autonomous agents, self-healing knowledge graphs, or why

"Earth orbits Pasta" is still the best bug report I've ever received.