swarmclawai/swarmvault

The local-first LLM Wiki: open-source knowledge graph builder, RAG knowledge base, and agent memory store. Built on Andrej Karpathy's pattern. An Obsidian alternative for personal knowledge management, AI second brain, and durable Claude Code / Codex / OpenClaw memory.

What it solves

SwarmVault provides a local-first way to organize fragmented information—such as documents, code, transcripts, and URLs—into a structured, durable markdown wiki and a machine-readable knowledge graph. It solves the problem of "knowledge scatter" by automating the maintenance of connections between sources, preventing the manual effort typically required to build a personal knowledge base or a RAG (Retrieval-Augmented Generation) system for AI agents.

How it works

SwarmVault employs a three-layer architecture to manage data:

  1. Raw sources: Immutable copies of original documents (books, code, etc.) are stored in a raw/ directory.
  2. The wiki: LLM-generated and human-authored markdown pages (summaries, entity pages, cross-references) are stored in a wiki/ directory.
  3. The schema: A swarmvault.schema.md file defines the structure and conventions the LLM follows when organizing the wiki.

Users can ingest data via a CLI, which then compiles the information into a graph. The system supports hybrid search (combining SQLite full-text search with semantic embeddings) and provides a graph viewer for navigation. It can run entirely offline using a heuristic provider or local LLMs via Ollama, or connect to cloud APIs.

Who it’s for

  • Researchers and students conducting deep-dives or creating book companions.
  • Developers building code documentation or managing project knowledge.
  • Knowledge workers performing business intelligence or personal knowledge management.
  • AI Agent developers who need a structured, graph-based knowledge base to provide context to agents via MCP servers or context packs.

Highlights

  • Local-first & Offline: Works without API keys using a built-in heuristic provider or local LLMs (e.g., Ollama).
  • Graph-based Navigation: Includes an interactive graph viewer and commands to query paths and explain relationships.
  • Hallucination Control: Tags edges as extracted, inferred, or ambiguous and includes automatic contradiction detection.
  • Agent Integration: Supports various AI agents (Claude, Cursor, Copilot, etc.) via hooks and an MCP server.
  • Multi-format Ingest: Handles 30+ input formats, including GitHub repos, YouTube transcripts, and local audio files via Whisper.