NoTrack Coder 2.8.61 is out, and it adds the QA Reviewer: after the agent writes code, a second model of your choice reviews what it just changed and points out problems — so you stop copy-pasting your code into another AI for a sanity check. It's off by default; you pick a mode and a reviewer model.

What the QA Reviewer does

After an agent answer, a second model reviews the code the agent just wrote and reports back either PASS — looks good or a short list of concrete problems. If there are problems, you click Apply fixes and the agent reviews and fixes exactly those files, or Dismiss to ignore them. Nothing changes on its own — fixes are applied only when you click.

Three modes

  • Off — no review; the agent works exactly as before.
  • Suggest — after each answer a card shows a Start review button; the review runs only when you click it, so it only costs anything when you ask.
  • Auto — the review runs automatically after each answer and shows the result.

Pick your reviewer

The reviewer can be any OpenAI-compatible endpoint you've added (for example your own OpenAI key with gpt-4o) or the built-in notrack-coder-uncensored model — self-review with no external key and no external provider. With an external endpoint the review runs on your key and your cost, and only the changed code is sent there. If an external reviewer refuses on content-policy grounds, the card tells you to switch to a different reviewer model — the built-in model handles those cases, so you're never left blocked.

Private by design

Only the code the agent changed that turn is sent to the reviewer — not your prompt and not the chat. Use the built-in model and the whole review stays on your machine, in keeping with how the rest of NoTrack Coder works: local, no telemetry.

How to set it up and use it

First, update to 2.8.61 (existing installs auto-update). Then:

1. (Optional) Add your own OpenAI model as the reviewer

  • Settings → Cloud Models / Endpoints → Add
  • Base URL: https://api.openai.com/v1
  • API key: your OpenAI key
  • Model: e.g. gpt-4o
  • Save — the model is now available to pick as a reviewer.

Prefer not to add anything? Skip this step and use the built-in notrack-coder-uncensored model (no key, nothing leaves your machine).

2. Turn the reviewer on

  • Settings → Agent tab → QA Reviewer (right under Desktop control).
  • Mode: Suggest (a Start review button appears after each answer — runs on click) or Auto (runs automatically after each answer); Off disables it.
  • Reviewer model: your OpenAI model from step 1, or the built-in notrack-coder-uncensored.

3. How it works in a chat

  • After the agent writes code, the reviewer checks only the code the agent changed this turn — not your prompt, not the chat.
  • It returns PASS — looks good or a short list of concrete problems.
  • If there are problems, click Apply fixes and the agent reviews and fixes exactly those files, or Dismiss to ignore them. Nothing changes on its own — fixes apply only on your click.

If an external reviewer refuses on content-policy grounds, the card tells you to pick a different reviewer model — the built-in model handles those cases, so you're never left blocked.

Get 2.8.61

Download for Windows, or just let your current install auto-update. See the plans if you're not on a paid tier yet.

Frequently asked questions

What is the QA Reviewer in NoTrack Coder?

A second model of your choice that reviews the code the agent just wrote, after each answer, and reports problems. It's built into NoTrack Coder from version 2.8.61 and is off by default — you turn it on and pick a reviewer model.

Does the QA Reviewer change my code automatically?

No. It only reports PASS or a short list of problems; fixes are sent to the agent only when you click Apply fixes. Nothing happens on its own.

What gets sent to the reviewer?

Only the code the agent changed that turn — not your prompt and not the chat. With the built-in model the review stays on your own machine.

Which models can act as the reviewer?

Any OpenAI-compatible endpoint you've added (your own key, your cost) or the built-in notrack-coder-uncensored model, which lets you self-review with no external key or provider.

What if the reviewer refuses on content grounds?

The card tells you to pick a different reviewer model. The built-in uncensored model handles those cases, so a content-policy refusal never leaves you stuck.