Which nodes
It picks the roles the task needs — Investigate, Plan, Create — and drops them on the canvas.
Back to all features
knowledge + extensibility · prompt → graph
Model Prompt.
Write a prompt; the model builds the graph. Hand the model full control and a live read of the canvas — it decides which nodes to place, how to wire them, and which model and tools each one runs, adapting as the graph grows.
Testing
prompt → graph
Full model control
what it is
One prompt becomes a graph.
Describe the outcome you want in plain language. The model reads your prompt and the live canvas, then builds the node graph for you — placing nodes, drawing the edges between them, and assigning a model and tools to each.
It reasons about the work the way you would: what to investigate first, where to fan out in Parallel, where a Merge pulls the threads back together, and where a Review gate belongs. As the graph grows it keeps a dynamic understanding of everything already on the canvas, so each new node wires into the structure that's already there instead of starting from scratch.
You stay in command: every node it places is a real, editable node. Accept the graph as-is, retarget a model, or extend the prompt and watch it grow.
placement
wiring
How they connect
It draws the edges so output and control flow downstream in the right order.
assignment
Model + tools per node
Each node gets a fitting model and the tools it's allowed to call — a small model where it's cheap, a strong one where it counts.
awareness
Live canvas read
It sees what's already placed and wires new work into it instead of duplicating nodes.
structure
Branch + gate
It introduces Parallel, Merge, and Review gates when the work calls for them.
adaptation
Grows with the prompt
Extend the prompt and the graph extends — new nodes attach to the existing chain.
two directions
Prompt and graph, both ways.
A graph can be compiled down into one portable prompt. Model Prompt runs the opposite way — it reads a prompt and builds the graph. Same bridge between the two, opposite directions.
model prompt
Testing
prompt graph
You write what you want. The model decides the nodes, the wiring, and the model/tools per node — and keeps a live read of the canvas as it builds.
graph → prompt
In development
graph prompt
The opposite of compiling a graph down into a prompt — an existing canvas reduces to one portable, engineered prompt you can paste into any model.
interactive · the builder
Pick a prompt. Watch the graph build.
Choose a prompt below, then press Build graph. The model reasons step by step, places each node, and wires it into a runnable chain with marching edges.
your prompt
Testing
Pick a prompt — the model builds a graph from it.
› Research 3 competitors and draft a one-page brief.
model reasoning
canvas · model-built
ready
model-driven · editable nodes
0 nodes · 0 edges
One prompt → a wired graph. Every node it places is real and editable on your canvas.
how it works
From prompt to wired chain.
The model treats the canvas as a tool it can act on, building the graph in a controlled loop.
01
Read prompt + canvas
It parses your intent and takes a live read of what's already on the canvas — existing nodes, edges, and context.
02
Plan + place nodes
It reasons about the steps, then places each node and assigns its model and tools — fanning out and gating where the work calls for it.
03
Wire + hand back
It draws the edges into a runnable chain and hands you an editable graph — press Run, or keep extending the prompt.
related features
In developmentMODELS.md
Directive Model Control
One version-controlled file governs every model on the canvas.
openActiveVoiceOrb
Agentic Voice Control
Talk to your canvas — it edits the graph as you describe it.
openActiveper node
Model Hotswapping
Set provider + model per node and swap live, mid-run.
open