- Category
- AI Agents
- Rank
- No. 106Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- @garrytan
- GitHub
- 29.7k stars
- Latest release
- v0.48.5.0
- Date
About
A personal knowledge management system that creates a searchable, AI-powered "memex" from your markdown files, notes, meetings, and calendar data. Uses Postgres + pgvector for hybrid search and integrates with AI agents to continuously enrich and maintain your knowledge base.
What it does
GBrain turns stored material into cited answers instead of returning a reading list. It gathers relevant pages, facts, and graph relationships, then synthesizes an answer and identifies missing or stale context. Typed links connect entities such as people, companies, meetings, and investments. It can run locally, connect to coding agents through MCP, or operate continuously with a hosted agent.
Why it's ranked here
The combination of answer synthesis, citations, gap reporting, and graph traversal is technically substantial. The repository also exposes a broad CLI, MCP transport, library exports, two storage engines, migrations, and extensive verification scripts. The strongest performance claims come from a synthetic benchmark in a separate repository, so they support interest rather than settling real-world quality.
What's good
It remains useful without API keys through keyword search and agent-written memory. Adding a supported provider enables semantic search and automatic fact extraction. Local PGLite avoids a database server, while Postgres provides the remote, concurrent path. Search implementations return a common result shape, with fusion and deduplication shared above the storage layer. Synthesized answers retain citations and explicitly report knowledge gaps.
Tradeoffs
The easy local engine is single-process, single-machine, manually backed up, and recommended for fewer than 1,000 files. Remote, concurrent use requires Postgres infrastructure. Semantic retrieval and extraction need an external model key. Continuous enrichment needs an always-on agent, a server with at least 8GB RAM, and API spending. The synthesis pipeline documents only its first round as fully exercised, while later rounds repeat retrieval without specialized gap filling.
How to use it well
Use it when an agent needs durable context for meeting preparation, relationship history, research, or institutional memory. Start locally with keyword search, then add embeddings when semantic recall justifies external API use. Move to Postgres for teams, multiple devices, or concurrent access. Treat citations and gap notices as prompts to verify stale material. It does not replace source connectors, managed backups, or a system that guarantees every email and message was captured.
Technical notes+
package.json defines a Bun and TypeScript ESM package, the gbrain CLI binary, public library subpaths, compiled macOS ARM64 and Linux x64 builds, and MIT licensing. src/cli.ts dispatches a large operation registry plus many CLI-only commands. src/mcp/server.ts exposes operations over MCP stdio, applies surface filtering and source scoping, and adds PGLite IPC context services. src/core/think/index.ts implements gather, synthesize, citation resolution, gap output, optional persistence, model routing, and usage accounting. docs/ENGINES.md describes a shared BrainEngine contract implemented by embedded PGLite and Postgres, with RRF fusion and deduplication above engine-specific keyword and vector search.
Observed
- License
- MIT
- Primary language
- TypeScript
- Runtime
- Bun 1.3.10 or newer
- Installation surface
- Installed from GitHub with Bun, or cloned and linked locally; the npm package with the same name is unrelated
- Interfaces
- CLI, TypeScript library exports, and MCP stdio server
- Storage engines
- Embedded PGLite and Postgres with pgvector
- Compiled targets
- macOS ARM64 and Linux x64 build scripts
Read from README.md, package.json, src/cli.ts, src/core/index.ts, src/mcp/server.ts, src/core/think/index.ts, src/core/minions/index.ts, src/core/artifact/index.ts, src/core/ingestion/index.ts, src/core/resolvers/index.ts, src/core/remediation/index.ts, src/core/schema-pack/index.ts, src/core/distribution/index.ts, src/commands/migrations/index.ts, docs/ENGINES.md.
What it can do
Ingest and index markdown files
Markdown files → Searchable knowledge base entries
Process and store meeting data
Meeting recordings or transcripts → Structured meeting notes and searchable content
Import calendar events and context
Calendar data → Timestamped knowledge entries with event context
Perform hybrid search across knowledge base
Search query → Ranked relevant documents and notes
Generate AI-powered content enrichment
Existing notes and documents → Enhanced content with additional insights and connections
Create semantic connections between documents
Multiple related documents or notes → Knowledge graph with linked concepts and relationships
Run automated knowledge maintenance cycles
Existing knowledge base → Updated and refined knowledge entries
Intel on GBrain
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.