Graft just crossed 1,600 GitHub stars. Open-source context layer for coding agents, works with Codex too.
Heat trend
The percentage is based on available heat signal, not comment count or independent people.
This covers a coding tool or code-capability update — useful for developers assessing workflow changes and reusable value.
Built this because I kept watching AI coding agents re-explore the same codebase every session, same files re-read, same imports re-followed, nothing carried over from the last run.
Graft writes a map of your codebase into linked markdown files, committed to git, so the agent reads that instead of starting from zero. Structural layer is tree-sitter, no LLM required, works on any repo size. If you want LLM-written summaries on top, it's provider-agnostic, so it works with OpenAI's API directly, not tied to any one model.
Works with Codex through MCP and an AGENTS.md integration, plus a post-edit hook that keeps the graph in sync automatically after every edit.
Ran a 162-task benchmark: 46% fewer tool calls, 42% fewer tokens, 60% less time, correctness held steady.