Skip to content

Workflow, Part 1 — Modes

Modes change how the LLM behaves depending on what you want to do. Instead of one fixed behavior for every conversation, a mode lets you switch the model between, say, thinking carefully and planning versus actually writing code.

warpdrv ships with two built-in modes:

  • Plan — the model investigates and researches, then proposes a plan. It makes no edits and waits for you to approve the plan first.
  • Build — the model strictly implements the plan you agreed on, and stops to ask before doing anything out of scope.

Modes are a great way to separate planning from development, or to dedicate a mode to research. Each mode has a color that’s shown in the UI so you can always tell which mode you’re in.

A mode is more than a label. Each one can define:

  • Its own list of tools — the only tools the model may use in that mode.
  • Its own agents — the subagents it may spawn (see Workflow, Part 3 — Subagents).
  • Its own guardrails — safety checks applied to the work (see Workflow, Part 2 — Guardrails).
  • A customized prompt — instructions that tell the AI how to operate while in this mode.

warpdrv rejects any tool call that isn’t in the active mode’s allowed list. So if the model tries to call a tool the mode doesn’t permit, the call is refused. This means a mode can’t accidentally use a tool it shouldn’t — for example, a read-only planning mode can’t write files.

  • Create a mode with the /create mode slash command.
  • Edit a mode from the right-hand panel on the Chat page. There you can change a mode’s tools, agents, guardrails, and prompt.

Use the mode tabs in the UI to switch between modes. You can also set a default mode for a workspace so it’s applied automatically to new threads.

Switching between two non-default modes does not reprocess the prompt. However, turning a mode off (switching to the default mode) or switching from the default mode to a mode does trigger a reprocessing of the entire prompt.