# Guide the Agent

Give it the right sources, choose how it works, and return to a creative run.

Source: https://www.creativly.ai/docs/agent/context-and-controls

![Agent composer with one blue-cup image selected, an inline aurora-launch-directions note mention, and a request naming Nano Banana 2.](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-selected-context-53d0294ab0679399.webp)

A selected image above the message, a named note inside it, and a clear request. This is a draft—not a submitted generation.

For an edit, say **which source to use**, **what should change**, and **what must stay**. Check the selected-item chips before sending; an unrelated selection can point the Agent at the wrong source.

## Bring the right context

### Canvas & notes

Click an image or file on the canvas—or in its **Assets** panel—to select it. It appears above the message under **Selected**. Remove a selection you do not want included.

Type **@** or choose **Elements** to name an available note or canvas. Select the suggestion rather than just typing a filename. Here, the selected image is the source and the mentioned note provides creative direction.

![Typing @ in Agent opens a real picker for creator-guide and aurora-launch-directions notes, while an image remains selected.](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-mention-picker-998f27f3a29027b3.webp)

### Files & tools

Open **+ Add context**. **Upload file** brings in new media or a document; **Canvas assets** opens existing work. **Connectors** brings up connected tools, and **Skills** adds a reusable approach.

![Actual Add context menu with Upload file, Take photo, Canvas assets, Connectors, Skills and Elements, above the Agent composer controls.](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-add-context-6e24f4e3adeecbef.webp)

Check the attachment chips before sending. The files upload to the canvas when you send the message. If an upload fails, resolve that error first; typing a filename does not attach the file.

[Use a skill with /](https://www.creativly.ai/docs/agent/skills) · [Connect your tools](https://www.creativly.ai/docs/agent/connections)

## Review an image against a brief

Use **+ Add context → Upload file** to attach an image and a PDF brief. Ask a specific question and say if you only want feedback—not new media.

### Attach & ask

![Agent composer with a blue-cup image, PDF attachment icon, and a request to check the studio still against the Aurora brief without making new media.](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-attachment-draft-1e5df06e1bd2d97d.webp)

An image and a PDF attached to the same message. The request asks for a short review, not a generation.

### Read the review

![Submitted image-and-PDF message and Agent's completed review: the blue cup fits the campaign, but sharper sunlight would better match the brief's studio direction.](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-attachment-review-4781dfec2fb23262.webp)

The actual reply connects the image to the brief and suggests harder, more directional light.

The uploaded files remain on the canvas for reuse. This example produced feedback only; it did not edit the image or generate another one.

### Try the same example

Download the [studio image](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/flow-video-still-original-697b14be7527665e.png) and [campaign brief](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-example-brief-d4fe54496ed4d497.pdf), then attach both.

> Does this studio still fit the attached Aurora brief? Give me a short verdict and one specific improvement. Just review the two attached files—don't generate media, edit the canvas or create any files.

## Choose how it works

### Thinking or Direct

**Thinking** gives the Agent room to ask clarifying questions. **Direct** moves ahead with fewer check-ins. Choose the mode beside the send button; neither choice selects an image or video model.

![Agent mode menu with Thinking checked and Direct available above the composer.](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-thinking-direct-bc517b3561d785d3.webp)

Want advice before making anything? Say so: **“Let's choose the direction first. Don't generate media yet.”**

### Chat model

The model selector at the **top of chat** chooses the assistant that plans and coordinates the work. It is separate from the **media model** that makes an image or video.

![Actual conversation-model selector with Luna selected and other OpenAI, Google, Creativly Agent and Claude choices, each with a relative cost indicator.](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-conversation-model-c47f224e300be664.webp)

Choose the assistant here. Name a media model in your request when that choice matters.

For example, **Luna** can handle the conversation while **Nano Banana 2** creates the image. Check the generation card's model label to see which media model was actually used. Available choices can differ by account.

## Answer a question, then continue

When a question appears, select an option or type your own answer, then choose **Confirm**. Selecting an option alone does not continue the run. **Skip** continues without supplying that answer.

![Agent clarification card asking which feeling Aurora should own, with Calm Studio selected, Sunny Coastal available, a custom-answer field, Skip and Confirm.](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-question-answer-0d5f6ff4122c8be7.webp)

A real planning-only run: choose Calm Studio, then Confirm. The Agent returned a direction without generating another image.

## Return to earlier work

Click the **conversation title or back arrow at the top of chat** to open **Creative Runs**, then choose the earlier conversation. Its messages and results reopen in the same workspace.

![Creative Runs list with the completed planning conversation under Now and Aurora Coffee Launch Directions under Today.](https://pub-60e30bed45514f03933d3900731bea31.r2.dev/public-app-sessions/v1/documentation/agent-creative-runs-b3d8b720f136d55e.webp)

The planning conversation and the earlier four-image launch brief are separate Creative Runs.

For a new brief, return to this list and write a new message. To continue an earlier brief, open its run first. **Follow** controls whether the canvas follows the Agent's work; it does not choose the source media.

For media work, check **Ready** on the result card and open the actual output. A finished reply or completed conversation is not a substitute for reviewing an image, playing a video or downloading a document.

Open in Creativly: [Agent ↗](https://www.creativly.ai/agent)