# Agent Settings

Source: [Agent Settings](<https://pr-995-preview.chatcut.dev/docs/agent-settings>)

Documentation index: [llms.txt](<https://pr-995-preview.chatcut.dev/docs/llms.txt>)

Last updated: 2026-11-19

Agent Settings controls how the current Agent works and which generation confirmations ChatCut may skip. Where you find it depends on the app:

-   **On the web:** the sliders button labeled **Agent settings** beside **Agent** at the lower-left of the Prompt composer.
-   **In ChatCut Desktop:** the **Settings** section of the **Agent Selection** menu. Desktop has no separate sliders button.

## Open Agent Settings

On the web:

1.  Open the **AI** panel.
2.  Make sure the composer is in **Agent** mode.
3.  Select the sliders button labeled **Agent settings** beside the mode selector.

The button is not the **Video settings** control used in Video Gen mode.

In ChatCut Desktop:

1.  Open a project and make sure the composer is in **Agent** mode.
2.  Click the agent button next to the mode selector at the bottom of the AI composer.
3.  Look under **Settings**, below the list of agents. The options shown there belong to the agent selected in **Agent Selection**.

## Model and reasoning options

The built-in ChatCut Agent chooses its model and reasoning effort for you, so there are no model controls for it.

In ChatCut Desktop, an agent you run on your own account can expose its own options under **Settings**. For example, **Codex CLI** shows **Model**, **Effort**, and **Speed**, and **Claude Code** shows **Model** and, when Claude Code provides it, **Effort**. If an option is absent, the agent manages it automatically. See [ChatCut Desktop](https://pr-995-preview.chatcut.dev/docs/desktop-app) to choose an agent.

ChatCut Desktop also has **Settings → Generation**, where **Default generation models** sets the image and video models used when a conversation has no model of its own. Leave either one on **Automatic (AI chooses)** to let ChatCut pick.

## Motion Graphics Quality

On the web, choose **Speed**, **Balance**, or **Quality** under **Motion Graphics Quality**:

-   **Speed:** fastest iteration and simpler results.
-   **Balance:** the default tradeoff between speed and quality.
-   **Quality:** best results and the slowest option.

ChatCut Desktop has no Motion Graphics Quality menu. There, the Agent writes motion graphics in the project itself; see [Motion Graphics](https://pr-995-preview.chatcut.dev/docs/motion-graphics).

## Auto-allow generation

On the web, open **Auto-allow generation** in Agent settings and turn each type on or off: **AI Avatar Generation**, **Motion Graphics**, **Video Generation**, **Video Translation**, **Voice Generation**, and **Image Generation**. The **Motion Graphics** choice applies to all your projects; the others apply to the current project. Motion Graphics and Voice Generation start switched on; the other types ask until you allow them.

In ChatCut Desktop, open a project, click the agent button in the AI composer, and under the **Settings** heading of the **Agent Selection** menu, open **Auto-allow generation**. Choose **Ask every time**, **Auto-allow**, or **Custom**. **Custom** lists AI Avatar Video, Image Generation, Video Generation, Video Translation, and Voice Generation separately. The choice belongs to the current project, and a new project starts from the choice you made most recently.

Turning a type off restores the matching confirmation card. Lip-sync changes always ask. This setting controls approval, not whether the action uses credits or whether the current account can access the feature.

The confirmation card itself is explained under [Generation Confirmations and Auto-Allow](https://pr-995-preview.chatcut.dev/docs/generation-confirmations).

## How does the Agent remember things?

ChatCut can keep a short memory of how you work, so you do not have to repeat yourself. It remembers things like who you make videos for, how a series should end, or a correction you gave once. Memory is shared across the web app, ChatCut Desktop, and connected agents, and it follows your account rather than one project.

Depending on your account, memory may start switched on or off until you choose. To check it, open the **Memory** page and look at the **Use memory** switch:

-   **On the web:** open the avatar menu, choose **Settings**, then **Memory**.
-   **In ChatCut Desktop:** open Settings, then **Memory**.

When **Use memory** is off, the Agent cannot read saved memories and does not save new ones. If you ask it to remember something, it tells you to turn memory on first.

There is no list to browse or edit by hand. You manage memories by talking to the Agent. After it saves, changes, or removes a memory, it tells you in one sentence what it did and which videos the memory applies to, so you can widen, narrow, or remove it.

-   **Save something:** say “remember that my product videos always end on the logo.”
-   **Check:** ask “what do you remember about my captions?”
-   **Remove something:** say “forget my caption preference.”

The Agent saves at the reach your words support. A one-off instruction stays in the current chat. “From now on” or “always” becomes a default for what you named. It does not save credentials, other people’s private details, or guesses about your taste. Memory never overrides your settings or paid-generation approvals.

Prompt

```text
Remember that my vlog captions should be one size smaller than the default, for all my vlogs.
```
