uncmatteth/gbrain-bridgebrain

BridgeBrain is an installer and local service that connects GBrain (an AI agent memory system) to text embeddings without requiring a GPU, Ollama, or a separate API key for an embedding provider. Instead, it piggybacks on the user's existing Codex CLI login to ChatGPT, using that already-authenticated session as the source of the embedding capability. Embeddings are numerical representations of text that let AI systems search and compare content by meaning rather than exact word matching. GBrain needs them to store and retrieve useful context for AI agents. The service runs entirely on your local machine, listening on a loopback address (127.0.0.1) that nothing outside your computer can reach. It presents an API that looks like OpenAI's embedding API, which means GBrain can talk to it without knowing it is actually routing through your ChatGPT login. Text passed through the service is converted to a 1536-dimensional vector and cached by a hash of the source text, so repeated lookups do not re-process the same content. Raw source text is not stored in the cache, only the derived vectors. The default installation puts GBrain in "side brain" mode: it gains the embedding endpoint, can search and store context better, but does not automatically scan your whole machine or rewire other AI tools like Cursor or OpenClaw. Upgrading to "main brain" mode, where GBrain is hooked into those other agents and given access to specific folders on your computer, is a separate and explicit step that requires setting environment variables and specifying which directories to sync. The README is written with security boundaries spelled out clearly. The installer stops if Codex is not already logged in rather than attempting to handle credentials itself. It never copies, prints, or persists raw authentication data. A dry-run mode lets you see exactly what the installer would do before it writes any files or modifies anything. The repo works on Linux, macOS, and Windows and includes service installers for systemd, launchd, and Windows Scheduled Tasks so the embedding service can start automatically at login. It also ships test fixtures and an evaluation harness for measuring how accurately the embeddings support search and recall tasks.