Graft is the code-mapping tool that finally stops AI coding assistants from wasting your tokens on repetitive codebase exploration.

@github.signals

4.3K followers · usually 5.8K views a post

Views

41.6K

Likes

1.8K

Comments

440

SoundTikTok sound · original sound - github.signals

Promoted product

Graft

GitHub is where people build software. More than 150 million people use GitHub to discover, fork, and contribute to over 420 million projects.

Why it worked

The video effectively highlights a common pain point for developers using AI coding assistants (wasted tokens and repetitive exploration) and presents Graft as a clear, efficient solution. The use of benchmark data and a concise explanation of the problem and solution likely resonated with the target audience.

Summary

Graft is a code-mapping tool designed to prevent AI coding assistants from wasting tokens on repetitive codebase exploration. It builds a permanent local graph of your codebase, which can be shared and regenerated by teammates, improving efficiency and reducing overhead.

Structure

1Introduction to Graft and its purpose
2Explanation of the problem with current AI agents
3How Graft solves the problem by building a local graph
4Details on sharing and regenerating the graph

Original caption

Your coding agent keeps exploring the same repository from scratch—and you are paying for it every time. Graft is the code-mapping tool that finally stops AI coding assistants from wasting your tokens on repetitive codebase exploration. Whenever an AI agent starts a task, it blindly searches your files to build a mental map that it immediately throws away when the session ends. Graft solves this by building a permanent, local graph of your codebase once, saving it as a folder of simple markdown files right in your repository. It integrates directly with tools like Claude Code and Cursor to instantly feed them the exact context they need, making your AI runs up to three times faster and four times cheaper. Best of all, it automatically updates itself... Open source: nanonets/graft Spotted on Hacker News. Follow for more open source project drops. #ClaudeCode #CodingTok #AICoding

More from @github.signals

On-screen text

Graft Turbocharge Claude Code, Cursor, Codex, Gemini & every coding agent: faster, cheaper, with contextual understanding specific to your codebase. WEBSITE GRAFT.NANONETS.AI NPM v0.8.1 DOWNLOADS 3.2K/MONTH NODE >20 TYPESCRIPT STRICT LICENSE MIT TELEMETRY NONE OPENSSF SCORECARD 6.7 Up to 4x cheaper and 3x faster, with better or no loss of correctness. vs. a standard session, no Graft With Graft Tool calls 48% fewer Tokens 42% fewer Time 62% less Correctness equal The table is a 102-run controlled benchmark (same agent, same file tools, only the context differs). The "Up to 4x cheaper / 3x faster" figures are the biggest single-task wins from a separate real-repo sweep (PocketBase, Ollama, Excalidraw). Full method & per-repo numbers npm install -g @nanonetshq/graft graft init # install the CLI, once # build the graph & wire it into # Claude Welcome back. # Graft is active. # Build prompt: worst beta Claude code # Graft prompt: worst beta Claude code # Claude # "Where does capture happen?" # Graft # 482 nodes / 857 edges. 79% enriched. - export Contents Quick start The problem What Graft does How the graph gets built What's in a node What runs where Agent integration - MCP server - Claude Code (deep integration) CLI Search & orient (graft grep / graft map) Monorepos & multi-repo folders Visualize it (graft viz) Benchmark Tested on your popular repos Development License Quick start npm install -g @nanonetshq/graft graft init # install the CLI, once # build the graph & wire it into That is the whole setup. graft init: asks which of your coding agents to wire up, builds graft/ from your code, and drops a statusline and hooks into .claude/. So from the next session on Graft rides along in Claude Code: It pulls the matching nodes into each prompt and rebuilds the graph in the background after every turn. No daemon, no re-indexing to remember, nothing to run or maintain by default—the graph is just files. Nothing is written until you pick. Run graft init --dry-run to see every file it would touch first, or graft init --agents claude to skip the prompt and wire Claude Code alone. graft build adds graft/ to your .gitignore automatically—the graph is a local, regenerable cache (like node_modules), not something you commit. What you share is the wiring that dropped into .claude/, each teammate runs graft build to generate their own graph. git add --claude && git commit -m "wire in graft" Prefer not to install globally? npx @nanonetshq/graft init works the same way. The problem Every task, your coding agent starts blind. Before it changes anything, it re-explores the repo: grep a term, open a file, follow an import, back out, try again. It is rebuilding a picture of a codebase it mapped an hour ago and threw away. That rediscovery burns most of a run's tool calls, tokens, and latency, and it is pure overhead. Discarded. Whatever the agent figured out dies with the session.