AI Providers

Return offers two ways to run AI: locally on your machine or through Return’s cloud servers.

Local mode

Local mode runs all AI requests through Return’s built-in engine on your computer. Nothing leaves your machine - your documents stay completely private. This is the default mode and works without an account or subscription.

You choose which model to use (like Qwen or Bielik) and can switch models at any time. See Local AI (Built-in Engine) for model recommendations.

Best for: Privacy, offline work, zero cost.

Cloud mode

Cloud mode sends AI requests through Return’s servers to Anthropic, whose larger, more capable models answer them. Return’s servers pass your text on and do not store it. There’s no setup - just switch the mode and start working.

Cloud mode requires a Pro subscription. Each cloud request counts as an action, and your remaining action quota is shown in the AI settings.

Best for: Powerful analysis, complex tasks, no local hardware requirements.

When a cloud answer is cut off or declined

Every answer has a length limit. A cloud model that thinks before it answers gets extra room for that, but a long answer can still reach the limit. Return never passes off a cut answer as a whole one:

• Chat asks the model once more with more room (that request is an action too). If the answer still does not fit, you get what came, marked as cut off: ask it to continue, or narrow the question.
• A text Return shows what came, with a note under it. A Return that would change your document leaves the text as it is and offers Retry, and so does any other Return whose answer was cut off.
• A large document is analyzed in sections, and a section whose answer was cut off counts as failed: the result says how many sections could not be analyzed and offers Retry failed sections, which runs just those again. In Playbook Check, the positions a cut answer did not reach show as Not checked.

The cloud model can also decline a request, usually because of what the text is about. Return then shows a message instead of an answer. In chat, rephrasing may help. A Return would most likely be declined again, so run it with the local model, where nothing leaves your computer.

A cloud answer that stops before you get anything from it (cut off while the model was still thinking, or declined) does not count against your monthly actions, within a small daily allowance.

Switching modes

Open Settings (Cmd+,), go to the AI tab, and toggle between Local and Cloud. The switch takes effect immediately - your next message will use the selected mode.

Your chat history is preserved when switching. You can go back and forth freely, for example using local mode for everyday editing and cloud mode when you need deeper analysis.

Comparing modes