fastercode

Make your Python faster, with every rewrite checked against the original.

Your AI agent (Claude Code or Codex) writes the rewrites. fastercode checks every one against your original, measures it, and only offers you a rewrite that passed every check and is really faster. Nothing changes in your code unless you choose Apply.

Two ways to use it

• Auto mode: turn on Fastest first, while your agent writes in the goatedcode sidebar, and your agent writes the fastest correct version first, as it codes.
• Refactor: click Make faster above any function to speed up code that already exists, one function at a time, with the full check below.

How a refactor works

1. Pick a function. Click Make faster above any Python function.
2. Choose the goal: speed, or memory.
3. Your agent takes its turn in its own workspace: it profiles the function, tries drafts, and fastercode checks each draft as it goes. Your files aren't touched while it works.
4. fastercode runs the official check, independently:
   • the rewrite against your original on recorded and generated inputs;
   • edge cases and side effects (files, output, arguments changed);
   • your project's own tests;
   • calibrated timing, so "faster" is never just noise.
5. You decide. You get the result, a report and a diff. Apply writes the winner into your file.

The goatedcode fuzzing method

Every rewrite is checked by the goatedcode engine, not by the agent that wrote it:

• Your real inputs: the calls your program actually makes, recorded and replayed against the original and the rewrite.
• Generated edge cases: fuzzed inputs (empty, zero, negative, huge, unusual values), plus a search for any input where the two disagree.
• Side effects: files written, output printed and arguments changed, compared as well as return values.
• Your own tests: the project's tests that call the code are run again.

The tests are saved locally. Every check becomes an ordinary test script on your machine, in goatedcode's own folder, or in your repo's tests/goated/ if you choose. You can read it, run it and add your own cases, and goatedcode includes them in every later check.

The reports are saved locally too. Every run's report (what was checked, what changed and the measurements) is kept on your machine, in goatedcode's own folder outside your repo, so you can open it again or attach it to a pull request.

Your code stays yours

Everything runs on your machine. fastercode never uploads your code.

Part of goatedcode

fastercode is one of the goatedcode tools, built on the same engine as simplercode, with more coming. Installing it also installs the goatedcode core, which sets up its own engine on first use (you need Python 3.10 or newer; your project's environment is never touched). Free in VS Code while we launch. See fastercode.ai and goatedcode.com.