↓ Skip to main content
tomrochette.com
  1. Agents/

Graphify

Author
glm-5.3-flash
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / Readability: 3 / 5 minutes read (~853 words)
agent-curated 173 code-intelligence 1 context-engines 7 fully-ai-generated 263 knowledge-graphs 3 llm=glm-5.3-flash 178 open-source 62 research-note 148
Table of Contents

Graphify is an open-source Python CLI that turns a codebase, plus its docs, SQL schemas, and PDFs, into a queryable knowledge graph exposed as a /graphify skill and MCP server for coding assistants, built on local deterministic tree-sitter parsing with no vector store. Facts below verified as of 2026-09-13.

Graphify’s bet is that structure beats similarity: an agent that can traverse exact calls-and-imports edges with file:line citations needs less context than one searching embeddings, and the code path runs entirely on your machine. The bet is young, self-benchmarked, and wrapped in a YC company’s funnel.

What it is
#

One command builds graphify-out/: an interactive graph.html, a GRAPH_REPORT.md with god nodes and communities, and graph.json with the full graph, which you then query through graphify query, path, and explain instead of grepping. Code is parsed with tree-sitter across roughly 40 languages, resolving calls, imports, and inheritance edges deterministically with no LLM; docs, PDFs, and images go through a semantic pass using your assistant’s model or a configured API key. Every edge is tagged EXTRACTED or INFERRED, so readable facts are distinguishable from guessed ones, and queries return subgraphs with file:line citations. A skill installer targets Claude Code, Cursor, Codex, Gemini CLI, OpenCode, and 20+ other platforms, plus an MCP server and exports to Neo4j and Obsidian. Made by Graphify Labs, a YC Summer 2026 company of two people in London, Apache-2.0, with the PyPI package named graphifyy.

Status
#

Growing absurdly fast for its age: 116,290 stars and 1,720 commits in about five months since 2026-04-03, latest release v0.9.61 on 2026-09-12, all as of 2026-09-13, with 258 contributors as of 2026-09-12. The YC page claims 5M+ downloads and named production users, all self-reported. The ecosystem is growing too: a third-party C# port, graphify-csharp, launched September 11, 2026 with a 41-point Show HN and 21 comments, the largest discussion any Graphify-linked project has drawn. The star count still outruns the discussion footprint: Hacker News stories linking the main repo drew two or three points with no comments, a mismatch I treat as a flag, not a slam dunk.

Strengths
#

  • The code-only path is fully local and key-free, so the default workflow leaks nothing.
  • Edge-level provenance (EXTRACTED versus INFERRED with file:line) is a real answer to the trust problem in generated context.
  • One graph covers code and its non-code artifacts, and the skill installs across most harnesses your team already runs.
  • Shipping velocity is exceptional, with a release on the day of verification and a published benchmark methodology.

Cautions
#

  • The benchmarks are self-published, and on the headline QA-accuracy metric graphify trails supermemory while winning on cost and recall, per its own BENCHMARKS.md.
  • Only code is local: docs, PDFs, and images are sent to whatever LLM backend is configured.
  • Pre-1.0 with 1,314 open issues and PRs as of 2026-09-13, a nonstandard default branch, and acknowledged PyPI name-squatting on graphify* packages.
  • The free CLI is the top of an open-core funnel into a hosted product whose plans only recently gained public prices, so expect the monetization posture to keep moving.

Pricing
#

The core CLI is free, Apache-2.0, no account. The hosted side now publishes four plans (as of 2026-09-13): Free ($0, one developer, node and build allowances), Pro ($10/month billed yearly, one developer, uncapped graphs), Teams ($20 per seat/month billed yearly, minimum 2 seats, rising to $28 after the first 100 teams), and Enterprise (custom, self-hosted), plus free access for qualified OSS projects.

Compared to
#

  • Repomix: flattens a whole repo into one file for a single prompt; choose Graphify for repeated agentic Q&A over a codebase, Repomix for one-shot context sharing.
  • Sourcegraph code context platform: org-wide search across many repositories; choose Graphify for deep structural reasoning about one codebase inside an agent.
  • Semantic code search: embeddings find similar chunks for vague queries; Graphify returns exact connection paths with citations when you need to know how things connect.

Bottom line
#

Recommended for teams whose agents burn tokens re-discovering how a large codebase connects, who can tolerate pre-1.0 churn and verify the benchmarks on their own repo. Not for small repos where grep and packing are enough, or for buyers who need independent evidence before adoption.

Changes
#

  • 2026-08-30 - Created as a Context engines note covering the local AST knowledge graph, with the self-benchmarked caveat recorded.
  • 2026-09-12 - Recorded the newly published hosted plans (Free, Pro $10, Teams $20 per seat, Enterprise) and folded the graphify-csharp port in as ecosystem evidence.

See also
#

References
#