GBrain (garrytan/gbrain, 30,000+ GitHub stars when checked on 2026-10-08) validates the idea of persistent AI memory: a database your agents read and write over MCP, local PGLite by default (GBrain README). Brainfile is the same idea — without the infrastructure tax.
Free by email: a practical Claude Code tip every Tuesday and Thursday, plus a Sunday recap.
GBrain: an open-source MIT-licensed memory system for the agents you already use, Claude Code, Codex, Cursor and others (garrytan/gbrain, 30,000+ GitHub stars on 2026-10-08). By default it keeps memories in a local PGLite database with keyword retrieval and needs no API key; semantic search through an embedding provider (Voyage by default, or OpenAI, Gemini, Ollama and others) and a Postgres server are optional, per its README.
Brainfile: a library of maintained CLAUDE.md configurations — structured role and project context that Claude Code reads automatically from your project root, updated as Claude evolves.
Both aim to give Claude standing context instead of re-explaining yourself every session. GBrain does it with a database and semantic retrieval; Brainfile does it with a plain markdown file Claude reads on its own, no infrastructure required.
bun install -g github:garrytan/gbraingbrain init --pglite --no-embeddingclaude mcp add gbrain)[PLACEHOLDER] sections with your info| Feature | GBrain | Brainfile |
|---|---|---|
| AI model support | Any MCP client (Claude Code, Codex, Cursor and others); embeddings optional, many providers | Claude Code (CLAUDE.md files); the free kit's paste-in configurations also work in ChatGPT and Gemini |
| Infrastructure required | Bun 1.4+ and a local PGLite database by default; Postgres + pgvector optional | None — just your existing Claude subscription |
| Claude Code integration | Native: a Claude Code plugin and an MCP server | Native — configurations are CLAUDE.md files, work immediately |
| Setup time | Install, init, connect (README says about 30 minutes) | Pull the files, run /start to fill them in |
| Monthly infrastructure cost | None for a keyless local brain; embedding and enrichment providers bill separately | No additional cost — included in your subscription |
| Profession-specific configurations | General-purpose memory; your agents write it as they work | 15 expert-crafted configurations (Developer, Founder, Trader, Marketer, and 11 more) |
| Long-term maintenance | You maintain the infrastructure, keep dependencies updated | We maintain the files; a weekly freshness feed tracks Claude Code changes |
| Works offline / local | Runs locally on PGLite; your agent still calls its model provider | Plain files on your machine; Claude Code still calls Anthropic's API |
| Licensing | MIT (open source) | Commercial — active development, email support and a weekly freshness feed |
| Context approach | Vector embeddings (semantic search over stored memories) | Structured markdown (explicit, transparent, fully controllable) |
You're a developer who enjoys building and maintaining your own infrastructure, want one memory several agents share, with sources and corrections, and semantic search across thousands of stored facts with full control over every part of the system.
You want to spend time on your actual work, not setting up databases. You use Claude Code and want a CLAUDE.md that works immediately, with expert-crafted configurations you can customize in 30 minutes and updates as Claude evolves.
GBrain's vector memory and Brainfile's CLAUDE.md solve different problems — semantic recall across ad-hoc notes versus a maintained role-context file Claude reads automatically every session. If you already run GBrain's infrastructure, a CLAUDE.md costs nothing extra to add alongside it. Start with the 3 free configurations and see whether the persistent half is worth it.
GBrain uses vector embeddings — it stores memories as semantic chunks and retrieves relevant ones automatically based on your query. This is powerful for unstructured data at scale.
Brainfile uses structured markdown — everything is explicit, transparent, and fully in your control. You read it, you edit it, you know exactly what Claude sees. No black-box retrieval, no wondering why it did not include something.
The Autopilot protocol, 10 pattern files and a weekly freshness feed. Pro adds 15 role packs. Works with your existing Claude subscription.
30-day money-back guarantee · Instant access · Cancel anytime
AI context engineering tips and new configuration releases. Free: a Claude Code tip every Tuesday and Thursday, plus a Sunday recap.
No spam. Unsubscribe anytime.