Gobed as the search engine

Gobed brings static embedding search to Go, which lets GPU Brain keep retrieval close to the web server, CLI, and MCP tools.

lee101/gobed is the Go search layer behind GPU Brain's embedding path. It wraps static embeddings for Go applications and includes CPU, int8, and GPU-oriented search work.

In GPU Brain, gobed handles local chunk embeddings while PostgreSQL stores pages, tags, links, timelines, keyword indexes, and pgvector rows. The product surface stays simple: agents ask for memory, not vector math.

Current production shape

• Markdown pages remain the source of truth.
• Chunks are embedded locally and stored next to page metadata.
• Keyword recall and vector recall can be fused for better practical search.
• REST, CLI, hosted MCP, and local MCP use the same page and search contracts.

Why this matters

The search engine can evolve without changing how agents use memory. Today the hosted app uses the Go service and PostgreSQL path. The same contract leaves room for larger GPU indexes as workloads grow.