---
title: "Choose review models | Proofline Help"
description: "Choose the models used to generate hypotheses and assess evidence in the review settings. Configure provider access before selecting a model that needs it."
url: https://proofline.sh/help/review-models/
---

[Help center](/help/)

# Choose review models

Choose the models used to generate hypotheses and assess evidence in the review settings. Configure provider access before selecting a model that needs it.

## Configure provider access

Open **Settings → Review policy → Models**, select **Change** under **Review model**, and open the **Providers** tab. Select a provider to add, replace or delete its key, or to connect a custom provider. Complete the provider setup required for your organization's chosen models.

Keep provider credentials in the connection controls. Do not paste them into review rules, repository configuration, or finding comments.

## Choose models

1.  Open **Settings → Review policy → Models**.
2.  Under **Review model**, select **Change** and choose the provider and model.
3.  On the **Hypothesis model** tab, choose another model, or keep **Same as review model**.
4.  When Proofline reviews deployments, choose the deployment models on the **Deployment reviews** tab, or keep **Same as review model** and **Same as hypothesis model**. A repository's **Settings → Models** page uses the same model names.
5.  Select **Done**, then **Save changes**.

**Repository overrides**, below the review model, lists each repository whose model choices replace these defaults.

Open the repository's **Settings → Models** page to change an override. The **Proofline default** summary includes its configured model name. If the page asks you to **Add an AI model provider**, add one on the organization's **AI model providers** page before overriding that default.

**Use organization default** removes one override; **Use organization defaults for all models** removes them all. Select **Save model settings** to apply the draft.

On the **Review model** tab, under **Sandbox harness**, keep **Automatic** to select a harness supported by the review model's provider, or choose **Claude Code** or **Codex**. **Claude Code** requires the Anthropic API or Claude Platform on AWS. Unsupported providers cannot run sandbox experiments. If the selected harness is incompatible with the review model, choose a compatible model or return to **Automatic**.

## Set credit limits

To bound how much work a review can spend, open **Settings → Review policy → Models** and set **Credits per review** under **Limits**. While it is empty, its note reads **Not set** and Proofline stops a review at its own spending ceiling. The review may still end incomplete within that budget.

Under **Limits**, set **Credits per pull request question** and **Questions per pull request per hour**. An answer that reaches its credit limit ends without a conclusion.

## Resolve unavailable options

If a provider or model is unavailable, check its configured connection before changing unrelated review settings.

A model choice does not expand the review's repository or environment access. An assessment still needs usable evidence for the environment.

## More tasks

-   [Restore a repository's organization model defaults](/help/restore-default-repository-model/)
-   [Choose separate models for deployment reviews](/help/choose-deployment-review-model/)
-   [Choose a sandbox harness for review experiments](/help/choose-review-sandbox-harness/)
-   [Replace a model provider's API key](/help/rotate-model-provider-key/)
-   [Test and connect a custom review model provider](/help/connect-custom-model-provider/)
-   [Limit the cost and frequency of pull request questions](/help/limit-pr-question-credits/)

## Related guides

-   [Configure review automation](/help/review-automation/)
-   [Review repository configuration](/help/repository-configuration/)
