What is gbrain?

GBrain is a long-term memory system for AI agents, always-on assistants that read emails, take notes from meetings, or follow your calendar. Out of the box those agents tend to be smart but forgetful: they answer the question in front of them and then lose everything once the conversation ends. GBrain gives them a "brain" that persists, so each new interaction can draw on what the agent has seen before. The brain ingests raw material, meetings, emails, tweets, voice calls, notes, into a local database. As each page is saved, GBrain extracts the people, companies, and other entities mentioned and creates typed links between them (such as "attended," "works_at," "invested_in," "founded," "advises") to build a knowledge graph, without extra calls to a language model. Retrieval uses a hybrid of vector search and that graph, so questions like "who works at Acme AI?" can be answered by following links, not similarity alone. Overnight jobs enrich entities, fix citations, and consolidate memory. The project ships 34 reusable skills, an evaluation harness called BrainBench, and an integration with the LongMemEval benchmark. You would set GBrain up if you run an autonomous AI agent, built on platforms like OpenClaw or Hermes, and want it to accumulate durable, structured memory about your work and contacts. Install takes about thirty minutes via an agent or a standalone command-line tool, the database is PGLite, which needs no server. It is written in TypeScript and runs under Bun, with built-in support for embedding providers including OpenAI, Voyage, Google Gemini, Azure OpenAI, Ollama, and llama.cpp. The full README is longer than what was provided.