How Agent Mode works
1
You describe what you need
Type your goal in the input bar. Attach files if you have reference material.
2
The agent plans its approach
The agent reads your message, loads relevant skills, and decides which tools to use. For complex tasks, it writes a plan in a scratchpad before starting.
3
The agent executes with tools
The agent calls tools as needed: writing code, running bash commands, generating images, searching the web, creating files. You see every action in real time.
4
Results stream to the side panel
Generated artifacts (websites, slides, images, videos, documents) appear in the side panel as they are created. You can interact with them immediately.
Model tiers
In Agent Mode, you select a tier rather than a specific model:
Select your tier from the model selector in the input bar before sending a message.
What the agent can do
Build
- Write and execute code (any language)
- Build and deploy websites
- Create presentations and reports
- Generate spreadsheets with formulas
- Create carousels for social media
Create media
- Generate images from text
- Create and edit videos
- Produce audio, music, and podcasts
- Remove and replace backgrounds
- Upscale images up to 4x
What you see in Agent Mode
Every tool call is visible in the chat:What you cannot do in Agent Mode
- You cannot pick a specific model. You choose a tier (Lite, Pro, Max) and the agent uses the corresponding model.
- You cannot limit which tools the agent uses. It decides which tools are appropriate for the task.
- You cannot run multiple agent tasks simultaneously in the same chat.
Next steps
Chat Mode
Switch to conversation-only mode for research and analysis.
Plan Mode
Make the agent plan before building for complex projects.
