Put several models in one ActionFlow

Most real jobs do not fit in one model call. They move through stages, and the stages want different strengths: a draft, a tighter extract, a format the downstream tool will accept.

ActionFlows lets those stages live in one ActionFlow. Each node can use a different model. The path still runs as a single run you can inspect.

Why one model is a weak default

A large reasoning model on a simple rewrite wastes time and credits. A fast model on a messy extraction drops fields. Locking every step to one vendor also makes the path brittle when that vendor changes a model name.

Composition on the canvas

On the canvas, models are nodes. You choose which model serves which step, then connect them so context moves forward. You can swap a model later without redrawing the whole job.

Tools finish the work

A model output that never leaves chat is still a draft. Integrations and MCP tools carry the result into the systems the team already uses. Multi-model composition matters most when the last step is a write, a ticket, or a file, not another paragraph.

A pattern that holds up

• Use one model to draft or explore.
• Use another to extract or structure.
• Use a later step to match tone or schema.
• Send the result through an integration when the job continues elsewhere.

This is not a party trick. It is matching tools to stages without losing the thread of the workflow.