# FAQ > Answers on cost, models, the reviewer's access, other reviewers, missing Codex login, privacy, the update check and resuming a stopped review. ## Cost and models ### What does a review cost? Each round is one turn of the reviewer model, billed to your Codex account (an OpenAI account or an API key). cross-review does not meter or show the cost. A review runs until the reviewer approves or the two sides stall, so the number of rounds depends on the change. ### Which model does the reviewer use? The reviewer’s own default model, with effort `high`, unless you choose otherwise. You can choose a model and effort in three places, from most to least specific: 1. In chat, when you ask for a review, for example “review it with a lower effort”. 2. In the [Settings](https://cross-review.tlelabs.com/docs/guides/settings/) page of the dashboard, or in a config file, per skill or for all skills. 3. Nowhere: the default applies. The runner remembers the model and effort of a session for its later rounds. ### Can the reviewer be something other than Codex? Not yet. Each reviewer CLI has one adapter in the runtime, and Codex is the only one. The `reviewer` setting accepts only the names the runtime knows. ## The reviewer ### Does the reviewer change my files? No. The review prompts forbid it, and the orchestrator makes every change. The reviewer runs with Codex’s full-access sandbox (`danger-full-access`), which lets it run your tests, builds and network commands such as `npm view` or `git fetch` to check a claim. That sandbox does not stop it from running anything else your account can run, so review the commands it reports, not only its findings. ### What does the reviewer send to the model provider? The prompt for each round, which includes the plan, the diff or the question, and the context the orchestrator passed. The reviewer also reads the files and commands it needs, and those reach its model as well. Codex talks to its own model provider, as it does for any Codex session. ### Why does a review say the reviewer is not found? The `codex` command is not in your `PATH`. The runner stops with the error code `REVIEWER_NOT_FOUND` and exit code 5. Install the Codex CLI with `npm install -g @openai/codex`, then start a new shell so the `PATH` change applies. ### Why does a review stop with a login error? The reviewer is installed but not logged in, or its login has expired. Run `codex login status` to check, and `codex login` to log in again. Then ask the agent to resume the stopped review; see [Can I resume a stopped review](#can-i-resume-a-stopped-review). > **Tip:** > > For the other causes of a stopped review, such as a stall, a timeout or a failed turn, see [Troubleshooting](https://cross-review.tlelabs.com/docs/guides/troubleshooting/). ## Privacy ### What stays on my machine? - Sessions, plans and settings stay in your repository’s `.cross-review/` folder and in `~/.cross-review/`. Nothing is uploaded by cross-review. - The dashboard listens on `127.0.0.1` only, and every request needs the token it prints. What leaves your machine is the reviewer’s traffic to its model provider, described above. ### What network request does the dashboard make? One. The update check reads the latest version of `@tlelabs/cross-review` from the npm registry. With **Check for updates** on, it runs when you turn the setting on and when the dashboard starts, at most once every three days; **Check now** runs it whenever you press it. The automatic check is off by default. Neither sends anything about you or your sessions. ## Resuming and changing a review ### Can I resume a stopped review? When the reviewer had started, the session has a reviewer thread, and the same session can usually continue. Ask the agent to resume it, for example “resume the stopped review in that session”. The agent sends its next reply to the same session with `review_reply`, so the reviewer keeps the earlier rounds. The agent does not start a new session on its own. If the reviewer never started, there is no thread to continue. Ask the agent to start a new review. ### Can I change the reviewer’s model in the middle of a review? Yes. Say so in chat, and the agent passes the new model or effort with its next reply. The new choice applies from that round on. ### Can I skip a review? Yes. Tell the agent to skip it, or to accept the current result. Your instruction overrides the skill; the agent records the approval as yours. See [Core concepts](https://cross-review.tlelabs.com/docs/start/concepts/#user-authority). ## Automatic reviews and plans ### Will reviews run without me asking? Only if you installed with `--auto`, and then only for `review-plan` after a plan and `review-impl` after an implementation. See [Automatic reviews](https://cross-review.tlelabs.com/docs/guides/automatic-reviews/). ### Are plans committed to git? No, by default. Turn on Plans in git to commit them. See [Plans](https://cross-review.tlelabs.com/docs/guides/plans/).