Graft: Persistent Codebase Memory Boosts Coding Agents' Speed & Efficiency
alex_verem · x · 2026-08-21
Open-source tool Graft stops coding agents from relearning your codebase every session. It writes a persistent map of linked Markdown files explaining the code structure, avoiding embeddings or vector DBs. Benchmarks on SWE-bench show a jump from 54% to 66% issue resolution, with 4x cheaper and 3x faster performance when agents use the map.
Related event: Graft Gives Coding Agents Persistent Memory of Codebases(2 posts)→
More from coding & agent
- Open Source Agent Harness: Turn READMEs into Social Media Posts — kmeanskaran · 2026-08-22
- ITSMBench Open Sourced: 50 Tasks for Multi-turn Agent Evaluation — Shahules786 · 2026-08-22
- Google Antigravity 2.9.1 ships faster startup, project switching and workspace polish — rseroter · 2026-08-22
- Seeking the Best Local AI Harness for Coding and General Agents — zyxciss · 2026-08-22
- Case Study: 80% of Report Agent Work Was Data Pipelines, Not AI — AmbassadorSad3889 · 2026-08-22
- AI coding agents cannot replace senior engineering thinking yet — bendee983 · 2026-08-22