Graft — Give coding agents a persistent, human-readable map of your codebase for faster, cheaper, and more accurate results.
Analyzed by Sai Pavan Gopularam · AI · Developer Tools · View on GitHub
- Stars: 6930
- Forks: 628
- Commits last 30 days: 100
- Health: Active (100 commits this month)
- Language: TypeScript
- License: MIT
What It Is
Graft is an open-source tool that builds a 'map' of your codebase for AI coding agents. Think of it like a senior engineer who has thoroughly documented every part of a complex system in plain English, explaining what each component does and how it connects to others, but for AI. It converts your code into a folder of linked markdown files, where each file describes a system, API, or concept.
This matters because AI agents usually start blind on every task, re-exploring the codebase from scratch, which wastes time, tokens, and money. Graft solves this by providing a persistent, always-fresh understanding of the code, allowing agents to skip the exploration phase and go straight to solving the problem, making them up to 4x cheaper and 3x faster with better accuracy.
License Verdict
MIT License — Build and Sell Freely — Commercial Use Approved • No Copyleft Restrictions
The MIT License is highly permissive. You can use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the software. This means you can build proprietary commercial products and services on top of Graft without needing to open-source your own code.
How to Use It
Graft sets up by installing a CLI tool globally or using `npx`. Then, you initialize it in your project, which builds the code graph and wires it into supported coding agents like Claude Code. The graph is stored locally and automatically rebuilds when code changes.
Prerequisites:
- Node.js 16+
- npm
Estimated setup time: 5 minutes.
npm install -g @nanonets/graft
graft init
git add .claude && git commit -m "wire in graft"What I'd Build With This
AI Agent Context Dashboard (micro-saas)
Build a web dashboard or VS Code extension that visualizes the Graft-generated code graph. Developers can browse nodes, see connections, and manually refine summaries or add notes. This helps them understand what their AI agents 'see' and allows for human-in-the-loop improvements to agent context. Charge individual developers or small teams a monthly subscription for enhanced visualization, graph editing UI, and integration with their preferred IDE/agent.
Effort: 2 Weeks Build Time · Target: Individual Devs, Small Dev Teams · Pricing: $19/mo per user
AI-Powered Codebase Onboarding & Knowledge Base (saas)
Create a SaaS platform that integrates with Git repositories (GitHub, GitLab). It automatically generates and maintains Graft graphs for all connected repos. New developers can use the platform to quickly onboard by querying the codebase through an AI interface powered by Graft's context. It also serves as a living, always-up-to-date internal knowledge base for engineers and agents. Offer team-based subscriptions with features like access control, versioning of graphs, and API access for custom agent integrations.
Effort: 3 Months Build Time · Target: Mid-sized Tech Companies · Pricing: $299/mo per team
Secure AI Dev Environment Suite (enterprise)
Develop an on-premise or private cloud solution for large enterprises. This suite would integrate Graft with their internal code repositories and security protocols, providing a secure environment for AI coding agents. It ensures sensitive code never leaves their network while agents benefit from Graft's efficiency. Offer custom integrations with existing enterprise tools, compliance reporting, and dedicated support. This is sold as an annual license or service contract.
Effort: 6 Months Build Time · Target: Large Enterprises, Regulated Industries · Pricing: $50k+/year
Sai Pavan Gopularam's Take
Graft is a smart way to make AI coding agents actually useful on real-world codebases. The problem it solves — agents starting blind every time — is a huge bottleneck. I'd lean into building a paid VS Code extension that makes managing and visualizing these code graphs a breeze, charging $19/month for individual developers who want their AI pair programmer to actually 'know' their code.
Watch Out For
- Telemetry Enabled by Default: Graft collects anonymous, opt-out telemetry data. While it's anonymous, some organizations might have policies against any form of data collection, requiring users to explicitly disable it via environment variables.
- LLM Costs for Deep Graph: Generating the full 'deep' graph with summaries and cruxes requires calls to an LLM provider (OpenAI, Anthropic, etc.). While structural graph building is free, the `--deep` option incurs token costs, which can add up for very large codebases or frequent full rebuilds.
- Local Cache, Not Committed: The generated `graft/` directory is a local cache and is automatically added to `.gitignore`. This means each developer needs to run `graft build` locally, and the graph itself isn't version-controlled alongside the source code.
I break down trending repos like Graft every week — join the newsletter.