GBrain — Open-Source Personal Knowledge Base

GBrain is a personal knowledge management (Knowledge Brain) tool inspired by Karpathy's LLMWiki style, which consolidates scattered markdown files into a Postgres-based knowledge base.

For hybrid search, it combines vector (pgvector HNSW) and keyword (tsvector) search using RRF (Reciprocal Rank Fusion), and improves search accuracy through a multi-query expansion pipeline leveraging Claude Haiku. Its knowledge model has a distinctive structure that separates compiled truth, which captures the current best understanding, from a timeline that tracks evidence.

It provides an MCP (Model Context Protocol) server with 20 tools, enabling direct integration with AI agents such as Claude Code and Cursor, and comes with the following 7 built-in AI agent skills:

• ingest: Document ingestion, knowledge rewriting, and cross-reference link generation
• query: Hybrid search and citation-based answer generation (hallucination prevention)
• maintain: Checking data health for contradictions, stale information, broken links, etc.
• enrich: Information enrichment via external APIs
• briefing: Organizing meeting context and deadlines
• migrate: Migrating data from various tools such as Obsidian, Notion, and Logseq
• install: Automating the installation process

For data processing, it uses a 3-stage chunking strategy (Recursive, Semantic, LLM-guided), and is provided as a TypeScript project under the MIT license.