Skip to content
CKS Documentation

The canonical knowledge backbone for LLMs

Verifiable, self-maintaining knowledge structures — validated against formal constraints, traceable to their source, and auditable end to end. No more citation hallucinations.
Get StartedOpen Interactive DemoView on GitHub

PyPI · cks-core

PyPI · cks-runtime

PyPI · cks-mcp

★ GitHub Stars

✓ 2,000+ tests across five repos

Terminal window
pip install cks-mcp

Installs cks-mcp, plus cks-runtime and cks-core as dependencies — one command, the whole backend. Then connect it to Claude Desktop with the 5‑minute Quick Start.

What’s inside

Section titled “What’s inside”

Five repositories, one canonical knowledge layer — a semantic engine at the bottom, an operational runtime on top of it, an MCP server that exposes it to LLMs, a visual workspace for humans, and this documentation site tying it together.

cks-core

The immutable semantic engine — canonical knowledge objects, validation against formal constraints, structural evolution. Docs → · Source ↗

cks-runtime

The operational environment — sessions, transactions, branching, three-way merges, and full version history on top of cks-core. Docs → · Source ↗

cks-mcp

The MCP server — exposes CKS to Claude Desktop and other MCP clients via 64 tools, plus MCP Resources and Prompts. Docs → · Source ↗

cks-studio

The visual workspace — a graph canvas, session gallery, agent pipeline monitor, and an in-browser AI chat panel that calls the same tools an LLM would. Docs → · Source ↗

Key capabilities

Section titled “Key capabilities”

No citation hallucinations

Mechanically detect references to non-existent sources before they ever reach a user.

Verification integrity

Cryptographic signing guarantees that source checks actually happened — not just that a field says so.

Full audit trail

Every operation is captured in an immutable version history you can list, diff, and roll back.

Branch, merge, sandbox

Isolate an experiment from the main line, reconcile it with a three-way merge, or throw it away.

Real semantic search

Find knowledge by meaning, not keywords — HuggingFace embeddings, or fully local with fastembed and no API keys.

LLM-native API

A native MCP server with 64 tools, fully compatible with Claude Desktop and other MCP clients.

How it works

Section titled “How it works”

An LLM and a human can drive the same knowledge graph at the same time — one over MCP through Claude Desktop, the other clicking around the graph canvas in cks-studio — because both go through the same cks-mcp tool surface into the same cks-runtime session.

Your LLM (Claude Desktop, etc.)      cks-studio (visual workspace)
        │                                     │
        └───────────────┬─────────────────────┘
                         ▼
                     cks-mcp ─── Model Context Protocol server
                         │
                         ▼
                  cks-runtime ─── Sessions, transactions, version history
                         │
                         ▼
                    cks-core ─── Immutable semantic engine

Why CKS?

Section titled “Why CKS?”

Today, the same knowledge exists in many incompatible forms — documents, databases, JSON, source code, AI prompts. CKS separates knowledge itself from every representation. Representations may change, but canonical knowledge remains the same.

Not sure where to start? The Quick Start guide gets you from pip install to your first validated knowledge structure in under five minutes, and the FAQ covers the questions that come up along the way.