Chat Workflows

The task thread is the main workspace for directing the agent. It combines prompt entry, file attachments, cloud media, model and runtime controls, streaming progress, memory context, and artifact review.

Start a Task

1. Open Home
2. Choose the workspace or project context
3. Enter a clear request in the composer
4. Attach files or cloud media when needed
5. Click the send button

Good prompts include the desired outcome, relevant constraints, and what the agent should avoid changing.

Attach Context

The composer supports multiple context sources:

• Local files
• Folders or workspace paths
• Cloud media from connected libraries
• Generated artifacts from earlier tasks
• Project context

Use attachments when the agent needs exact source material. Use plain text when the agent only needs instructions.

Edit and Regenerate

When a conversation path needs correction:

1. Edit the user message that caused the wrong direction
2. Regenerate the response
3. Compare the new branch with the old one
4. Continue from the branch that best matches your goal

Branching is useful when you want to test two approaches without losing the original task history.

Follow Streaming Progress

During execution, the task thread can show:

• Planning messages
• Tool use and status
• Generated files
• Memory context indicators
• Error messages and recovery prompts
• Output previews

If the agent asks for approval, open the approval prompt and decide whether it can continue.

Use Cloud Media in a Task

1. Open the attachment menu
2. Choose Cloud media
3. Select a connected provider
4. Search or browse folders
5. Preview the item
6. Attach selected media

Stock catalog attribution is added to the prompt when required. Personal media can include source details that help the agent publish edited results back to the correct destination.

Review Outputs

After the agent creates files or media:

1. Open the preview in the task thread
2. Check the content and metadata
3. Ask for a revision if needed
4. Export, attach to another task, or publish the final output

Do not publish or share generated output until you have inspected the preview and confirmed the destination.

Recover from Problems

• If the agent starts in the wrong direction, stop it and edit the prompt
• If a tool fails, ask the agent to explain the failure and suggest a fallback
• If an attachment is missing, reattach it and continue in the same thread
• If output is incomplete, ask for a revision with the missing criteria
• If the task should not proceed, reject the approval or stop the run