ADVANCED WORKFLOWS
Go further when you have a reason to. These workflows build on the basic guide; each points to the detailed setup instructions.
Separate parallel code changes#
Use New worktree when a task needs its own branch and working directory. Keep related work in the same thread; use separate threads and checkouts for independent changes.
On web and desktop, you can also Shift-click models in a new thread's model picker to send the same prompt to several models. Each selection starts a separate thread and worktree. This requires a Git project.
That gives you separate results to inspect; it doesn't choose a winner or combine the changes for you. Review the diffs before deciding what to keep.
Read threads and worktrees, source control, and worktree branch naming.
Switch providers with context#
You can continue a conversation with a different provider. T3 Code transfers conversation context when a native continuation isn't available.
Short conversations transfer intact. Longer ones use a selection of recent requests and answers, the original request, and relevant activity. A handoff doesn't copy the outgoing provider's reasoning, tool-call state, or attachments.
Repeat an important constraint in your next message when necessary. The saved thread history remains available through T3 Code's thread-reading tools.
Use separate provider instances when you need different accounts or configurations. Provider capabilities and permission behavior can differ.
Read context handoffs, provider setup, and permission modes.
Make projects fit your workflow#
Project overrides let a repository use different defaults from its environment. For example, a project can have its own model default, permission defaults, or worktree naming preferences.
In Settings, choose the project and environment the change applies to. The layers indicator shows whether a value comes from a built-in default, the environment, or a project override. Reset an override when you want it to inherit again.
Appearance and other device preferences remain local to the client. Providers and diagnostics are configured per machine.
Read settings and project overrides, model defaults, and appearance.
Schedule work and use webhooks#
Scheduled tasks run on the environment you choose, with their project, model, and workspace settings. On mobile, open Settings → Scheduled tasks to create or manage them. Fixed-time schedules use the environment's time zone.
A webhook task runs when its URL is called. Its prompt determines which request values the agent receives. For example, a task can receive a pull request URL from a webhook payload.
A public webhook URL needs a T3 Connect managed tunnel. Use the signature settings when connecting services such as GitHub, and test with a narrow task before broadening the work.
The host still needs to stay running and reachable for scheduled work. Read scheduled tasks and webhook automations before setting one up.
Connect tools and outside agents#
T3 Code's MCP server lets outside agents work with an environment: read projects and threads, start or message threads, and inspect available providers and models. Each outside agent signs in, and you choose what it may do.
Other tools let agents use browser sessions, create visual replies, or interact with connected devices. Start with a concrete workflow and check the relevant setup instructions and permissions.
Outside agents and MCP: connect your terminal agent or another MCP client.
Browser sessions: import a browser profile for browser work.
Visual replies: understand rendered HTML and MCP apps.
Devices: connect device viewers and tools.
These are optional capabilities. The basic project → thread → review workflow works without configuring them.