Choosing a mode
The mode menu is under the chat box, beside the model. A new chat starts in Build. The mode is per chat, and you can change it between messages.
Build
For doing the work. Describe what to build or change; the agent reads what it needs, edits files, runs commands and checks, and tells you what it did and what's left. Every change can be undone.
Good for: "add rate limiting to the login route", "migrate these components to the new design tokens", "make the test suite pass".
For truly big work (a redesign, a migration, a new subsystem), the agent offers a choice first: Plan it first, which switches to Architect, or Go straight to it. Ordinary changes go straight.
Ask
For questions. Ask looks wherever the answer is: this project, your other repositories, this computer and its programs, and the web. It reads, searches and runs read-only checks, and changes nothing. When you want the change made, switch to Build.
Good for: "how does auth work here?", "which services does this app call?", "what's using the disk on this machine?".
Architect
For planning before code. The agent reads the project and writes a plan you agree before anything changes: the structure, the files, the trade-offs and the risks. Change the plan in the chat; when it's right, have it built.
Good for: a new feature with moving parts, a refactor that touches many areas, choosing between approaches.
Debug
For a bug. The agent reproduces it, finds the cause, makes the smallest fix, and shows it's gone by running the same check again. Add clues as it works ("it only happens on Safari").
Good for: a failing test, a crash with a stack trace, "the checkout total is wrong".
Read-only setting
Settings, AI, Agent Mode off keeps the chat in Ask only. On (the default), all four modes are available.
Local models
Local models through Ollama work in every mode with the standard IDE tools. See Your own API keys and local models.