Open Brain (OB1)
A shared persistent memory layer for all your AI tools — one database, one AI gateway, one chat channel, any AI plugs in.
Evidence shows: security policy exists (SECURITY.md) but contact email is TODO placeholder; CI workflows use pull_request_target and secrets, least privilege insufficient; tests show API key passed via X-Brain-Key header, but no explicit user confirmation mechanism; data flow transparency limited, README does not detail data flows; sensitive data handling (e.g., importing Gmail, X data) mentioned but no specific security measures; dependency security not mentioned; external effects (e.g., deploying Edge Functions) documented but no rollback mechanism; source attribution clear (contributor list). Deductions: user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback all lack evidence or are missing.
Evidence shows: README and test files are self-consistent, tests cover multiple functions; dependency availability not explicit, but tests mock network; failure messages covered in tests for HTTP errors, but overall error handling documentation insufficient. Deductions: dependency availability evidence insufficient, failure messages only partially covered.
Evidence shows: README clearly defines target audience (non-coders) and multiple scenarios (extensions, recipes, skills); capability boundaries defined via extensions and recipes; trigger precision via step-by-step instructions, but no explicit automation triggers; environment fit (Supabase, K8s, Vercel, etc.) documented. Deductions: trigger precision evidence insufficient.
Evidence shows: information architecture clear (directory categories); install notes detailed (Setup Guide); naming stable (consistent directory naming); examples and FAQ exist; known limitations not explicitly listed; license is FSL-1.1-MIT, but metadata is NOASSERTION; versioning changelog not provided; maintenance responsibility clear (maintainer list). Deductions: known limitations, versioning changelog evidence insufficient.
Evidence shows: output usability demonstrated via README and tests; marginal value high (rich community contributions); cost-benefit not explicit, but deployment costs may be high. Deductions: cost-benefit evidence insufficient.
Evidence shows: README claims (e.g., 'build in 45 minutes') supported by docs, but not independently verified; cross-source corroboration limited (community contributors); facts and inferences not clearly separated. Deductions: claim traceability, cross-source corroboration, fact-inference separation all lack evidence.
- Security policy contact email is a TODO placeholder, needs to be filled.
- CI workflows use pull_request_target and secrets, posing potential risks; least privilege should be minimized.
- No explicit user confirmation mechanism; sensitive operations (e.g., data imports) may lack user consent.
- Dependency security not mentioned; dependency vulnerabilities should be checked.
- Rollback mechanism missing; deployment failures may not be recoverable.
What does this agent do, and when should you use it?
Open Brain is a self-hosted AI memory infrastructure that unifies your data into a single PostgreSQL database with vector search, and uses an AI gateway plus an MCP server to let Claude, ChatGPT, Cursor, and other AI tools share a persistent memory of you. It's not a notes app but an open protocol database, with quick capture via Slack or Discord. The system comprises a Supabase backend (database, vector embeddings, edge functions), an MCP server (e.g., remote MCP connection), and a rich set of community-built extensions, recipes, skills, and dashboards. You can set it up in about 45 minutes with the setup guide or an AI-assisted workflow, then build extensions in order, from a household knowledge base to a professional CRM, to enhance your AI assistant. The project is licensed under FSL-1.1-MIT, though the repository does not declare a license explicitly.
Open Brain runs a Supabase backend with a PostgreSQL database and pgvector for vector search, edge functions for AI inference, and an MCP server that connects AI clients (Claude, ChatGPT, Cursor) to the memory layer. The core data model is a 'thoughts' table, where entries captured via Slack/Discord or imported via data recipes (ChatGPT import, Obsidian vault import) are embedded and classified. Extensions (such as home maintenance tracker, professional CRM) and skill packs (like Panning for Gold, Auto-Capture) define outputs through SQL schemas and MCP tools. The deployment boundary is a self-hosted Supabase instance (or Kubernetes with PostgreSQL), and frontend dashboards can be hosted on Vercel/Netlify.
- A solo developer who wants Claude Code and ChatGPT to share project notes and avoid re-explaining context, using Open Brain as a unified memory layer.
- A knowledge worker who captures thoughts quickly via Slack and later asks an AI assistant to generate a daily digest of recent thoughts.
- A user migrating from Obsidian who uses the Obsidian Vault Import recipe to import notes with full frontmatter metadata preserved.
- A household user who builds the Family Calendar extension to coordinate multiple schedules, with the AI assistant reminding about events based on past memories.
- A job seeker who uses the Job Hunt Pipeline extension to track applications and interviews, and lets the AI assistant connect related notes and resume tips.
- An OpenClaw user who integrates Agent Memory to keep continuity across long-running tasks, persistent work logs across agents, models, and channels.
What are this agent's strengths and limitations?
- Provides a unified persistent memory layer so multiple AI tools (Claude, ChatGPT, Cursor) share context without redundant setup.
- Fully self-hosted; data stays in your own Supabase instance, and Kubernetes deployment removes the need for Supabase SaaS.
- Rich ecosystem of community contributions: extensions, recipes, skills, dashboards covering home, career, data import, and more.
- Built-in OpenClaw integration for persistent memory in agent workflows, including code review memory and task flow work logs.
- Supports many data import sources (ChatGPT, Obsidian, X, etc.) for migrating historical data.
- Initial setup involves multiple components (Supabase, Slack, MCP) and takes about 45 minutes, a steep learning curve for non-technical users.
- Relies on Supabase and pgvector; though Kubernetes offers self-hosting, there is some provider dependency.
- License is FSL-1.1-MIT but undocumented on GitHub (NOASSERTION), potentially complicating commercial usage.
- Many extensions/recipes, but quality varies and requires maintainer review; may be overwhelming to choose from.
- Community is English-centric; non-English users may find limited support.
How do you install or deploy this agent?
To install, follow the setup guide at docs/01-getting-started.md to build the full system (database, AI gateway, Slack capture, MCP server) in about 45 minutes, no coding experience needed; or watch the video walkthrough (vimeo link). Alternatively, use the AI-assisted setup (docs/04-ai-assisted-setup.md) and point Cursor or Claude Code at this repo to guide the build. You'll need a Supabase account (API keys and database credentials) and optionally a Slack app token for Slack capture. For MCP, configure your AI client to point to your remote MCP server URL.
How do you use this agent?
After installation, connect your AI clients (Claude Desktop, ChatGPT, Cursor) via the MCP server. Use the Slack or Discord bot to capture thoughts, which are automatically embedded and classified into the 'thoughts' table. Build extensions in order: start with Extension 1 and progress through 6 extensions, each README provides steps and expected outcomes. Use community recipes (e.g., ChatGPT import) to import historical data, and skill packs (like Panning for Gold) to extract ideas from brain dumps. Visualize data via dashboards like Open Brain Dashboard. For real-time help, join the Discord community.