What is Graphify? The code knowledge graph for AI agents

Graphify, a code knowledge graph for AI coding agents from Graphify Labs, now has an overview post explaining what the product is and how its two versions differ.

What it does

Graphify maps repositories into functions, classes, files, and the calls and imports between them. Agents such as Claude Code, Cursor and Codex query that map over MCP instead of grepping and re-reading files. It answers structural questions: who calls a function, what a module depends on, and what a change would reach. The agent gets a small set of connected nodes with file paths, so it reads the relevant code rather than whole directories.

How the graph is built

Graphify parses source with tree-sitter, links calls, imports and references across files, and groups related code into communities. Each edge carries a provenance tag: EXTRACTED when read directly from code, INFERRED when resolved, and AMBIGUOUS when there is more than one candidate. This lets an agent judge how far to trust a path.

Hosted or CLI

• Hosted Graphify (app.graphify.com): connect GitHub or GitLab and Graphify builds and updates the graph. Each push to the default branch starts a new build, and agents keep using the previous graph until the new one is ready. Agents connect to https://api.graphify.com/mcp and sign in with OAuth, so no API key is needed. The hosted app also covers cross-repository graphs, pull request reviews that read the graph, formal verification, a generated wiki, and shared agent memory. It is a commercial service with a Free plan, plus Pro and Teams plans.
• Graphify CLI: the open-source engine, published on PyPI as graphifyy, builds the same kind of graph locally. It supports about 40 languages, parses code locally, never executes source files, has no telemetry, and needs no model key for code-only graphs.

The hosted option suits code on GitHub or GitLab where the graph should rebuild on every push and be shared with a team. The CLI suits projects that live only on one machine or must stay offline.

Licence and data handling

The engine is open source under Apache 2.0. It was relicensed from MIT in July 2026, and the NOTICE file keeps MIT terms for portions contributed before then. Hosted builds run in isolated jobs with short-lived, repository-scoped credentials, and Graphify does not train models on customer code. The hosted service does send code to its LLM subprocessors, OpenAI and Anthropic, to produce results.

Limits and getting started

Graphify is strongest on multi-hop questions such as callers of callers and cross-repository dependencies. For an exact string like an error message, grep is still the better tool. Graphify is not a vector database, a chat memory product, or a replacement for code review, and the post advises checking cited source lines before relying on an answer.

To run the CLI locally, install it with uv tool install graphifyy and graphify install, then send /graphify . from an assistant inside a repository. The post also notes that the package name has two y's, and that other products called Graphify, including getgraphify.com and graphify.ai, are unrelated.