More than a box with a name.
Keep the full prompt and the input/output contract on the agent node. Read the contract on the canvas; open the prompt when you need the detail.
For the solo AI builder
Design your agent system.
One node at a time.
Give every agent a prompt, clear inputs and an expected output. Connect the pieces. Work out what depends on what—before you write the code.
A visual design tool. No model execution. No API keys.
THE UNIT OF DESIGN
Brief agentAgent
IN
source_notesOUT
evidence_briefPROMPT
Read {{source_notes}}. Write an evidence brief with observations, source labels, and unresolved questions.
One node. One responsibility.
A contract the next node can depend on.
A place to work out the architecture
The decisions belong beside the diagram.
Make the dependency visible.
Label what passes between nodes. Reconnect an input and rearrange the cards as your understanding of the system changes.
Carry the design into the build.
Export Markdown to brief yourself or a coding agent. Keep JSON for a portable backup of the prompts, connections and layout.
Deliberately focused
Design here.
Build and run elsewhere.
For the moment when a prompt chain has become too much to hold in your head—and you want to reason about it one node at a time.
In the app today
Visual authoring, node prompts and contracts, labeled connections, account saving, browser-only storage, JSON backups and Markdown handoffs.
Design before execution
Define the architecture here. Run models, evaluate prompts and implement the system in your development tools.
Your workspace
Sign in to save diagrams and folders to your account and open them on another device. Browser-only mode is also available.
Start smaller than the system
What does your first agent need?
Add one node. Write its input, prompt and expected output.
Then decide what comes next.
Sign in for account saving, or use browser-only mode.