Skip to main content
Agent chat is the fastest way to learn ADE — it shows the whole loop: prompt, tool calls, file changes, terminal output, review, and commit.

​
Start the chat

1

Start a new chat

In Work, open a new chat. You can pick an existing lane or let ADE create one for the task.
2

Pick a provider and model

Use the default, or choose any verified provider — Claude Code, Codex, Cursor, Factory Droid, or OpenCode — from the model selector.
3

Send a narrow task

Ask for one contained change. Skip broad prompts like “clean up the whole app” for your first run.
4

Watch tool calls

The transcript streams reads, edits, shell commands, and test output as the agent works.
5

Review the diff

Open Files, inspect every changed file, run tests if needed, then commit.
Start a new agent chat
Pick a provider and model for the chat
Choose an existing lane or let ADE auto-create one

​
First prompt

Add a GET /health endpoint that returns { "status": "ok" } and add one test for it.

​
If the agent gets stuck

  • Press Stop and ask for a shorter plan.
  • Paste the exact test or build error back into the chat.
  • Switch to a stronger model for the session.
  • Keep the task inside the lane — don’t move it to your primary checkout.

​
Next

Agent chat

How chats, providers, and tool calls work.

Pull requests

Open a PR once the lane is ready.