llmwiki

llmwiki compiles URLs, files, notes, research papers, documentation sites, and session exports into a persistent, citation-traced markdown wiki. It is a knowledge compiler rather than a retrieval-only system: sources are compiled once into a durable artifact, with concepts organized into typed pages and links forming a navigable graph.

The system uses a two-phase LLM pipeline that extracts concepts, merges duplicates across sources, and generates typed wiki pages with citations back to the original source lines. It supports configurable domain profiles through Configurable Lifecycle Profiles, which can define entity schemas, typed relations, lifecycle state machines, evidence gates, workflows, artifacts, connector bindings, content tiers, and retrieval policy. Projects without a profile use the classic concepts-and-queries behavior, and new projects can start with the built-in AutoSci research workflow or Newsroom editorial workflow. llmwiki also builds chunk-level embeddings for semantic search; query results are narrowed with cosine similarity, reranked with BM25, and expanded along the wiki graph to produce a citation-traceable evidence pack.

Delivery includes a local web viewer, the CLI, SDK access, and MCP integration. The viewer opens in a browser and provides sidebar navigation, full-text search, a force-directed page graph, and provenance chips on each paragraph. Through MCP, llmwiki can connect to Claude Desktop, Cursor, Claude Code, and other MCP-compatible agents so they can ingest sources, compile, query, lint, retrieve context packs, and exchange OKF bundles without using the CLI. The SDK offers in-process control of the pipeline, including OKF import and export. It also supports Google Cloud’s Open Knowledge Format initiative, with export to OKF bundles and import through a review queue while preserving producer metadata, llmwiki provenance, and safe original nested paths.

The documentation identifies AI researchers and engineers, technical writers, open-source maintainers, and developers as users. Installation requires Node.js 24 or higher and an LLM provider credential such as an Anthropic API key, an OpenAI-compatible endpoint, Ollama, GitHub Copilot, or a local Claude Code login.