graphCTX, a local memory system for AI coding agents, claims to retrieve repository context ~1ms compared with ~494ms for a cloud alternative, aiming to reduce the time developers spend re-explaining codebase details to AI assistants.

What happened: graphCTX is a locally-installed tool that captures and indexes coding facts—package scripts, lockfiles, CI configuration, and session details—and retrieves them in ~1ms per query. The system anchors memory to git commits and branches, promotes only durable knowledge, and ranks results to return only relevant context rather than a full memory dump.

Why it matters: Developers currently repeat themselves when working with AI coding agents because the agent lacks persistent, repo-specific memory. graphCTX aims to reduce that friction by storing and retrieving local context instantly, so developers spend less time re-explaining and more time shipping code. The system runs locally with no account, API key, or cloud setup required.

What to watch: graphCTX claims to maintain flat latency at scale—a 5,000-fact monorepo retrieves as fast as an empty one (p50 ~1.33ms), and it restored the needed repo fact in every run across 14 coding tasks after compaction. Installation takes three commands and works with Claude, Cursor, and other agents.