# Choose AI models

> Let Botchi choose each surface's model, or pin your own and review its price and privacy.

## Before you begin

AI model policies are available for business workspaces. Owners and dashboard editors can change them; viewers can inspect them. Policies apply to everyone in the account.

## How Automatic works

Every policy row has an **Automatic** switch, on until you choose your own. While it is on, Botchi chooses the model, reasoning level and provider for that surface and keeps them up to date when the Botchi default changes.

## Choose your own model

1. Open **Settings** and select **AI models**.
2. Open the policy row for the work you want to configure.
3. Turn off **Automatic**.
4. Select **Model** from the live managed catalog.
5. Select **Reasoning** if the model supports it.
6. Select **Provider** if the model offers more than one.
7. Select **Save**.

Saving pins the choice for that surface, even when it matches the current default, so later Botchi default updates no longer change it. Only enabled, available production choices appear. A **Custom** badge identifies an account override.

## Find the right policy

| Surface | Policies |
| --- | --- |
| Member chat | Conversations; subagents. |
| Account agents | Agent runs and chats; subagents. |
| Dashboard assistant | Assistant; read-only subagents. |
| Workflows & automations | Member automations; account workflows. |
| Panels | Panel reviews. |
| Apps | App builder chat. |

Each row is independent, including subagent policies.

## Read the facts

Each row shows, for the model and provider in use, zero data retention (ZDR), no-training coverage, inference region and price. Models without ZDR remain selectable.

For AWS Bedrock, the endpoint region is where requests enter AWS. Processing follows the exact inference profile: an EU profile stays in its verified EU destination regions, while a global profile can process outside the EU. Platform admins can inspect the invocation ID, destinations, current AWS pricing tiers and account quotas.

Context pricing thresholds come from official model documentation; the applicable tier prices the full request, including output and cache tokens. Unverified prices prevent model selection; a catalog listing alone does not confirm runtime IAM permission.

Routes displaying ZDR or no-training coverage enforce the corresponding filters across fallbacks. A route displaying no ZDR does not request ZDR. Platform-wide privacy settings can still apply independently.

Input and output multipliers compare base token prices with the automatic choice for that surface (1.00×). They do not assume an input/output mix.

> **Tip** — Compare both price multipliers and route privacy before saving. A lower input price alone does not predict the cost of a complete task.

## Return to Automatic

Turn **Automatic** back on for the policy. This removes your override and restores the Botchi default, so future default updates apply again.

## Related

- [Workspace settings](/en/docs/settings)
- [Sparks](/en/docs/concepts/sparks)
- [Review usage](/en/docs/usage)