SwarmVault
A self-hosted knowledge base that turns documents and code into a Markdown wiki and graph. MIT licensed, with offline workflows and optional Ollama support.
SwarmVault is a self-hosted knowledge base for people who want research, notes, or code documentation to become a persistent wiki that AI agents can use. It stores original sources, Markdown pages, and a knowledge graph as local files you can inspect. The project is open source under the MIT license.
Its built-in heuristic workflow works fully offline without API keys. For model-assisted extraction and synthesis, it supports local LLMs through Ollama, including Gemma. OpenAI, Anthropic, Gemini, and OpenAI-compatible APIs are optional cloud connections, so processing stays local when you use the offline workflow or a local backend.
The wiki follows naming rules and categories you define for your subject. It can turn books, papers, transcripts, screenshots, datasets, URLs, and repository code into linked source summaries, entity pages, and concept pages. Original source files remain unchanged, and graph connections record their supporting sources and claims.
Review controls let you inspect proposed changes before accepting them. SwarmVault flags conflicting claims, stale sources, missing citations, and coverage gaps. Search combines SQLite full-text matching with semantic embeddings when an embedding backend is available; graph views let you follow connections between pages.
For coding assistants and other AI agents, SwarmVault provides MCP access, integrations with Claude, Cursor, and Codex, and task history. A local browser viewer brings graph exploration, search, previews, and approval queues into one interface.
