Codistry ships a code map it calls experimental

Adronite's Codistry builds a local map of a repository's structure, then tells users not to trust it completely, per an August 2 review on WinBuzzer.

That matters for engineers who let an agent pick files for a task on its own. A wrong code graph will still call two files related, with total confidence.

What it does

Built by Adronite, the VS Code extension indexes a repository's structure locally. It finds code by concept instead of keyword, preselects files likely relevant to a task, and surfaces files that have historically changed together. The index refreshes after every commit, checkout, and pull.

The extension is model-agnostic. It connects directly to Anthropic and OpenAI, plus any OpenAI-compatible endpoint with tool calling. Users can also define subagents that carry their own constrained tools and max-turn limits.

Read past the feature list, and one detail stands out. Adronite's own documentation calls the index experimental, and tells users to treat its output as orientation, not truth.

It's the honest line for any AST-derived code graph, one built from static structure rather than a program actually running. Most vendors present their index results as settled fact instead of showing the seams. Say the seams out loud, and that becomes the reason to try Codistry before a competitor that won't.