Your own tools add real time

Running your own tools adds real wall-clock time that has nothing to do with the model's own reasoning speed. A bash call that runs your full test suite takes exactly as long as your test suite takes to run; tCode is simply waiting on the same process you would be waiting on if you had run it yourself.

A large repository being searched with grep across many files similarly takes longer than the identical search performed in a small one, for the same underlying reason: the search itself, not the model, is the slower part.

This means two requests that look similarly complex in their phrasing can take noticeably different amounts of time purely because of what they actually touch, a small utility script versus a large monorepo, independent of anything about how the request was worded.

This category of time is fundamentally outside tCode's control in the same way it would be outside yours if you ran the identical commands by hand; a slow test suite is slow regardless of who or what triggered it to run.

The fastest way to speed up a specific recurring kind of turn is often unrelated to tCode itself, a faster test runner, a narrower default search scope, the same optimizations that would help you personally running the same commands directly.