Graphify ranks 16 on Trendshift and crosses 131,000 stars for codebase graphs

A coding assistant that spends half its tokens re-reading files it already saw this morning is a cost line in a company's monthly bill. Graphify, an open-source tool from Graphify Labs, builds the map once and lets Claude Code, Cursor, Codex and Gemini CLI read from it on every follow-up question. The repository ranks 16 on Trendshift's daily list on 22 October 2026 and carries 131,795 stars on the GitHub API, up from zero on 3 April when the repo was created.

What the project describes

Graphify's README says the tool "turns any codebase, with its docs, SQL schemas, configs and PDFs, into a queryable knowledge graph". It uses tree-sitter to parse the source locally and emit an interactive HTML view, a Markdown report, and a JSON file. The repository calls itself a "/graphify skill" for Claude Code, Cursor, Codex and Gemini CLI, and it is Apache-2.0 licensed. Graphify Labs is a Y Combinator S26 company, which it states on its GitHub page.

The README states "cut agent token spend by 80 percent" as the value, and reports two benchmark numbers on the project's own code-understanding tests: 0.497 recall at 10 and 76 percent accuracy on a question-and-answer check. Both are Graphify Labs' own measurements. Graphify Labs says the parsing step uses no language-model tokens, because tree-sitter is a plain parser.

What it changes about a working session

The practical story is that an assistant that reads files on demand pays twice for anything it loads more than once. A static graph, written down and reused, is one of the cheapest ways to carry context between prompts. The README reports forty-plus languages and twenty-plus assistants, so the surface a team already has will almost certainly match. One cost that stays: the graph is only as current as the last /graphify run, and a long-lived branch will drift from a graph built at the top of the day.

What the number does and does not say

131,795 stars after six and a half months is a very large number for a developer tool. Trendshift's rank 16 is a daily place on a public list of what has gained attention in the last day, and that list is read by other developers, so the two numbers reinforce each other without being the same signal. The flip side is also true: a static graph is only useful to an authorised assistant working on code the team owns, not to anything running without permission.

The post does not say what a graph costs to build on a large repository, what the on-disk format looks like once a project has millions of lines, or how fresh the graph has to be before an assistant is giving a wrong answer from stale structure. A team weighing adoption will want its own number for a build of its own repository, and the open issues list on the project (1,612 open at the time of this post) suggests the shape and quality of that answer is still moving.

Source

GitHub: Graphify-Labs/graphify. Trending rank from Trendshift.