Gives AI agents persistent memory and a pre-built code dependency graph so they stop burning tokens on exploration. Two core pieces: hybrid search over decisions and patterns you've stored across sessions, and a tree-sitter powered graph that maps functions, classes, and call relationships across eight languages. Query what depends on a symbol or what breaks if you change it in under 1ms instead of spawning exploration agents. Backed by Postgres with pgvector for embeddings. The benchmarks on Click show 95% fewer tool calls when agents can ask the graph instead of grepping. Rust core, Docker Compose setup includes schema and model download. Works with Claude Code, Cursor, Codex, and any MCP client.
claude mcp add --transport stdio cdnsteve-remembrallmcp uvx remembrallmcp