Graft Boosts Claude Code Efficiency, Cuts Tokens by 42%
TL;DR. NanoNets introduces Graft, a new tool that optimizes coding agents like Claude Code by building a codebase understanding graph, leading to significant token and time savings. - Graft reduces Claude Code's token usage by 42% and increases correctness by 12 percentage points in SWE-bench tests by providing persistent context. - The tool integrates directly into development environments, allowing agents to access a pre-built graph of the codebase instead of re-exploring it each session. - It supports various coding agents and aims to make AI-assisted development faster and more cost-effective by eliminating redundant context generation.
- Graft reduces token usage by 42% and increases correctness by 12 points for Claude Code on SWE-bench.
- It works by building a persistent graph of a codebase, providing agents with pre-existing contextual understanding.
- The tool integrates with Claude Code, Cursor, Codex, and Gemini, improving speed and cost efficiency.
- Graft is designed to eliminate the need for coding agents to re-explore codebases from scratch in every session.