AI NewsOpen sourceAnnouncement
rust-anything hits 2,406 stars in two days, led by a coding agent
rust-anything turns a slow part of a TypeScript, Python or Go app into Rust. A coding agent does the port. The tool checks every step against the original.
Image: GitHub
Why it mattersA team with one slow endpoint no longer has to choose between a full rewrite and leaving it slow. The port is small, and a check catches any drift.
A new open-source tool called rust-anything has reached 2,406 stars on GitHub in two days. The repo was created by Erkam Yaman on 16 November 2026. It ranked first on Trendshift's daily list on 18 November. The pitch is short. Rewrite a slow part of an app in Rust, one piece at a time. A coding agent does the port. The tool runs a check against the original behaviour at every step.
The repository description is direct: "Rewrite your whole app in Rust, one piece at a time, checked against the original's recorded behavior at every step. CLIs, HTTP servers and libraries in TypeScript, Python, C, Go, Ruby and Java." PHP and C# are in preview. The licence is MIT.
How the check works
The idea is called behavioural equivalence. rust-anything runs the original code and records the inputs and outputs for each call. The README calls the record step "runs the original on generated edge cases ... and on every semantic trap in every parameter ... and saves cases.jsonl". The Rust port is then run against the same recorded cases. If any case differs, the port is wrong.
The workflow has seven commands: profile, bridge, record, build, diff, bench, gate. Profile picks the slow function. Bridge sets up the call path so the Rust version can be swapped in. Record captures behaviour. Build makes the Rust port. Diff and bench compare the two. Gate stops a release when the Rust version is not identical.
A coding agent drives it
The second piece is the MCP server. rust-anything runs as a stdio MCP server, called with rust-anything mcp. That means Claude Code, Codex, Cursor and Gemini CLI can each see the tool and call its commands. An agent reads the profile output and picks the hottest function. It runs the record step, writes the Rust, and runs the diff. It stops when the check passes.
The README shows one example number. On x18n's sidebar search over 50,000 keys, the Rust version took 0.5 ms against the TypeScript version at 10 ms. The author describes that as "20x faster". That is one function in one project on one machine. The same approach on another function will not land on the same figure. What the tool offers is the method of checking, which the figure helps show.
When this helps
A team with one slow endpoint has usually had two options. Rewrite the whole app in Rust, which is expensive and risky. Or add a cache around the slow call and live with it. rust-anything proposes a third option. Port only the hot function, keep every other line of the original, and prove the two match.
The limits are worth naming. The Rust port is still code the team has to own, debug, and build. A function that touches many services at once is harder to record than a pure function. And a behavioural check only catches what the recorder saw. An input the recorder never hit can still break production. For a well-isolated hot path in a TypeScript or Python app, this is a cleaner route than a full migration.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

