📊 See Graphify on AI Rank

📱 Get the AI Rank app

Graphify turns a codebase into a knowledge graph that a coding assistant can query instead of grepping through files. You type /graphify . in Claude Code (or Cursor, Codex, Gemini CLI and about 20 other assistants). It parses the code locally with tree-sitter, links functions, classes and imports across roughly 40 languages, and writes a graph your agent can search by question, path or concept. The open-source engine is Apache-2.0 and lives at Graphify-Labs/graphify on GitHub. A paid hosted version, Graphify Cloud, keeps the graph updated across repositories.

The repository was created on April 3, 2026 and had 132,244 stars and 12,679 forks on October 22, 2026. The Python package is graphifyy, with two y's, and version 0.9.93 shipped on October 20. Other graphify* packages on PyPI are not affiliated with the project.

What Graphify Builds From a Codebase

Code is parsed with tree-sitter ASTs, which is deterministic and needs no LLM. That part runs fully offline and nothing leaves the machine. Docs, PDFs and images get a separate semantic pass that uses the assistant's own model, or an API key you configure. Audio and video are transcribed locally with faster-whisper.

Each run writes three files to graphify-out/:

File What it holds
graph.html Interactive graph you open in a browser, with search, filters and clickable nodes
GRAPH_REPORT.md Most-connected "god nodes", detected communities, surprising links and suggested questions
graph.json The full graph, which the CLI and MCP server query later without re-reading files

Every edge is tagged EXTRACTED when it is explicit in the source, or INFERRED when Graphify resolved it. Communities come from Leiden clustering, and comments such as NOTE: and WHY: plus ADR or RFC references become their own nodes linked to the code. There are no embeddings and no vector store, so a query walks real relationships rather than similarity scores.

Graphify Claude Code Setup

Graphify needs Python 3.10 or newer and either uv or pipx. Install the CLI, then register the skill:

uv tool install graphifyy
graphify install --platform claude

A bare graphify install detects the assistant and defaults to Claude Code. To keep the skill inside one repository, so it can be committed, add --project:

graphify install --project --platform claude

Then open the project in Claude Code and type /graphify . in the chat. That is an assistant instruction, not a shell command. In Codex the skill is called with $graphify, and Codex needs multi_agent = true under [features] in ~/.codex/config.toml for parallel extraction.

On Claude Code, the integration also writes PreToolUse hooks. Before search-style tool calls, and before Claude reads source files one by one through Read or Glob, the hook nudges the agent toward the graph. Instruction-file platforms such as Codex, OpenCode and Cursor get the same guidance through AGENTS.md or .cursor/rules/. Remove everything with graphify uninstall, and add --purge to delete graphify-out/ as well.

Querying the Graph

After the first build, these commands run against graph.json from the repository root:

Command What it returns
graphify query "how does authentication reach the database?" A scoped subgraph for a plain-language question
graphify path "UserService" "DatabasePool" The shortest chain of edges between two nodes
graphify explain "RateLimiter" One node's source location, community and connections
/graphify . --update Re-extracts only changed files
graphify hook install Rebuilds on commit and branch checkout

If a name matches several nodes, Graphify lists the candidates instead of guessing. Exit code 0 means an answer, 3 an ambiguous name and 4 no matching node, which makes the CLI easy to script. A local MCP server with ten tools is available through the graphifyy[mcp] extra. Optional extras also add PDF, Office, SQL schema, live PostgreSQL, Terraform, Neo4j and Ollama support.

Privacy and Model Backends

A code-only graph needs no API key. For a mixed repository, --code-only skips the documents that would otherwise call a model. Headless graphify extract picks a backend from whichever key is set, in this order: Gemini, Kimi, Claude, OpenAI, DeepSeek, Azure, Bedrock, then Ollama. It prints which variable made the choice. Set GRAPHIFY_BACKEND=ollama to stay fully local, or GRAPHIFY_BACKEND=none to turn every LLM backend off. The README notes that the Kimi backend routes to Moonshot AI servers in China.

The project states it has no telemetry. It does log each query, path and explain call (question, corpus, node count, duration) to ~/.cache/graphify-queries.log. Set GRAPHIFY_QUERY_LOG_DISABLE=1 to turn that off.

Graphify Benchmarks and Popularity

The project's README reports these results. They come from Graphify's own harness, not an independent test:

Benchmark Metric Graphify Comparison in the README
LOCOMO (n=300) recall@10 0.497 mem0 0.048, supermemory 0.149
LOCOMO (n=300) QA accuracy 45.3% supermemory 49.7%, mem0 27.3%
LongMemEval-S (n=50) QA accuracy 76% tied with dense RAG
ERPNext cross-tool (n=6) key-fact coverage 82.0% grep/read baseline 70.8%

Graphify leads on retrieval recall but trails supermemory on LOCOMO answer accuracy, and two of the samples are small (50 and 6 tasks).

On AI Rank's GitHub repository ranking, Graphify was rank 15 with 130,990 stars and 4,591 new stars over seven days, listed under AI Engineering.

→ Track it on AI Rank: Graphify-Labs/graphify: stars, rank and weekly growth

Graphify Cloud Pricing

The open-source CLI is free. Graphify Cloud adds one live graph across repositories, PR review, formal verification runs and connectors for Sentry and Jira. Prices from graphify.com/pricing on October 22, 2026:

Plan Price Main limits
Free $0, one developer 3 repositories, 25,000 nodes per repo, 10 push builds per repo per day, 15 PR reviews per month
Pro $10/month billed yearly ($15 monthly) Unlimited repos, uncapped graph, 400 PR reviews and 400 verification runs per day
Teams $20/seat/month billed yearly for the first 100 teams, then $28 ($29 and $40 monthly); 2-seat minimum 700 PR reviews and 800 verification runs per day, shared graphs, Slack support
Enterprise Custom Self-hosted, SSO/SAML, GitHub Enterprise

Qualified non-commercial MIT or Apache projects can apply for free use. Pre-seed and seed startups under $1M revenue can get 50% off Teams for up to six months. Graphify markets Cloud as cutting agent token spend by 80%; that figure is the company's own claim.

Compare before you choose: star counts and rankings move daily. Check the current GitHub repository rankings on AI Rank before you commit.

Or 📱 get the AI Rank app to follow them on your phone.

Sources

Prepared with automated editorial assistance and checked on October 22, 2026 against the Graphify repository and README, the GitHub API, PyPI, Graphify's documentation and pricing page, and dated AI Rank repository data. We did not run our own benchmarks for this article.