Code Review Graph
An open-source project written in Python that builds a local-first, persistent knowledge graph for codebases. It helps AI programming tools accurately locate relevant context, reduce token consumption, and improve the efficiency of code review and large-repository workflows.
Timeline (last 90 days)
Code Review Graph is a free tool that claims it can reduce Claude Code's token usage by up to a factor of 49.
Code Review Graph filters files by building a code relationship graph (dependency graph), thereby reducing the number of tokens in the context.
According to Code Review Graph's own tests, in the best case token usage drops to 1/49 of the original.
On coding tasks, Code Review Graph saves 14 times the tokens on average; on code review tasks, it saves 6.8 times the tokens on average.
Code Review Graph's test data comes from the project's own testing, not from an independent third-party benchmark.
For small changes involving only a single file, Code Review Graph may not help, because there are already few files and reading the graph makes little difference.
Code Review Graph's installation method includes running code review graph install with pip, then executing build.
Code Review Graph is not exclusive to Claude Code; it is also compatible with tools such as Codex and Cursor.
Building and maintaining the relationship graph also has costs; when the codebase changes frequently, the graph needs to be rebuilt or incrementally updated, otherwise the graph becomes outdated and the basis for pruning becomes inaccurate.
All Facts (9)
Code Review Graph is a free tool that claims it can reduce Claude Code's token usage by up to a factor of 49.
50%UnverifiedCode Review Graph filters files by building a code relationship graph (dependency graph), thereby reducing the number of tokens in the context.
50%UnverifiedAccording to Code Review Graph's own tests, in the best case token usage drops to 1/49 of the original.
50%UnverifiedOn coding tasks, Code Review Graph saves 14 times the tokens on average; on code review tasks, it saves 6.8 times the tokens on average.
50%UnverifiedCode Review Graph's test data comes from the project's own testing, not from an independent third-party benchmark.
50%UnverifiedFor small changes involving only a single file, Code Review Graph may not help, because there are already few files and reading the graph makes little difference.
50%UnverifiedCode Review Graph's installation method includes running code review graph install with pip, then executing build.
50%UnverifiedCode Review Graph is not exclusive to Claude Code; it is also compatible with tools such as Codex and Cursor.
50%UnverifiedBuilding and maintaining the relationship graph also has costs; when the codebase changes frequently, the graph needs to be rebuilt or incrementally updated, otherwise the graph becomes outdated and the basis for pruning becomes inaccurate.
50%