Rewire Text 1.2.2: compare models without leaving the app

Rewire Text 1.2.2 is out, following a small 1.2.1 patch last week. Neither is a big release on its own, but the two additions in 1.2.2 combine into something useful if you care about which AI model does your work.

Model tier override. Every AI transform normally picks its own model tier: Light, Standard, or Advanced, depending on the job. A new tier control in the full window’s top bar (next to Settings) lets you take that decision over. Leave it on Auto and nothing changes. Lock it to a tier and every AI transform runs on that tier’s model until you switch back. The menu shows the model each tier currently resolves to on your provider, so it is always clear what you are switching between. The setting sticks across restarts, and the command palette honors it too. While a tier is locked, the control fills in so you can’t forget it is on.

Transform timing. The full window now reports how long each AI transform took, under the output box next to the character counts: elapsed time, the tier it ran at, and the exact model, e.g. 2.41s · Standard · gemini-3.5-flash. The timer counts up live while the result streams, so a slow model announces itself early.

Together, they make a nice model comparison setup. Say you are choosing between two models from your provider, hosted or local. Configure one as your Light model and the other as Standard, then use the tier control to flip between them. Run the same transform on the same text, read both results, and check the elapsed time under the output. A few rounds of that and you know which model is worth keeping, and where the trade-off between quality and speed sits for the transforms you actually use.

And the Windows 10 fix. 1.2.1 fixed a bug that could break transforms on Windows 10. If Rewire Text failed for you there, it works now. Windows 11 was never affected. The changelog has the details on both releases.

One more thing. A lot of people have taken the time to try Rewire Text these past few weeks, and many of you have written in with bug reports, suggestions, and the occasional kind word. Thank you. I read every message and respond to each one, and the app is better for it. If something is missing or almost-but-not-quite right, tell me.