Chirp

Chirp components are native Grasshopper nodes that run AI reasoning as part of your definition. They take data in, think it through, and pass structured results on — so a definition can make judgement calls, not just calculations.

What they’re for

Some steps in a design aren’t pure geometry — they’re decisions. Which option fits the brief? How should this be described? Is this result any good? Chirp lets you place that kind of reasoning right on the canvas, where it belongs, instead of breaking out to a separate tool.

You stay in control of the structure; Chirp handles the judgement inside the box you give it.

Categories

Each Chirp component plays a role you’d recognize from a design team:

Category	Role
Planner	Breaks a goal into steps
Interpreter	Turns loose input into structured data
Critic	Reviews a result and flags issues
Narrator	Describes what’s happening in plain language
Classifier	Sorts inputs into categories
Gate	Decides whether to pass or stop
Editor	Revises content against a goal

Cascades

Chain several Chirp components and they can share reasoning context — a cascade — so a multi-step design stays coherent from one decision to the next. Ask Rook to set one up by describing the chain you want.

Connect a model

Chirp calls a language model with your own key. Without a key, Chirp components still run: they return typed default values and show a warning, and any results you froze with the Freeze pin keep replaying.

1. Open %LOCALAPPDATA%\Rook\app\chirp\ in File Explorer.

2. Copy .env.example to a new file named .env, then replace your-key-here with your key:

   ANTHROPIC_API_KEY=sk-ant-...

3. Apply it. Chirp’s background process reads .env only when it starts, and it keeps running after you close Rhino. Sign out of Windows and back in (or, in Task Manager → Details, end the python.exe processes whose command line contains -m chirp --rook-managed). The next Chirp component you create starts it again with your key.

Set your key before you build Chirp components if you can.

Components you already have

Each Chirp component remembers the exact address of the background process it was created with, and every restart of that process (including a reboot) gets a new one. After a restart, existing components don’t reconnect when they recompute. They replay their frozen result or return typed defaults, with the warning adapter not running on localhost:…. To reconnect one, ask your agent to update the old port in that component’s script to the new one. That keeps its wires and its frozen results. Or ask it to recreate the component, then rewire it; a recreated component starts with no frozen results.

By default, planner components use anthropic/claude-opus-5 and the other categories use anthropic/claude-sonnet-5. To use another model or provider, set CHIRP_MODEL in the same file (it replaces both defaults), with that provider’s key:

CHIRP_MODEL starts with	Key
anthropic/	ANTHROPIC_API_KEY
openai/	OPENAI_API_KEY
openrouter/	OPENROUTER_API_KEY
gemini/	GEMINI_API_KEY

For another OpenAI-compatible endpoint, describe it in CHIRP_PROVIDERS:

CHIRP_MODEL=openai/my-model
CHIRP_PROVIDERS={"openai/my-model": {"api_base": "https://your-endpoint/v1", "api_key_env": "MY_ENDPOINT_KEY"}}
MY_ENDPOINT_KEY=...

Keep a copy of your .env

Installing a newer Rook over an existing one keeps your .env. Uninstalling Rook deletes it, together with the rest of the Rook app folder.

How you make one

You don’t configure Chirp by hand. Describe what the component should do:

“Add a critic that checks each layout against the brief and flags the weak ones.”

Rook places and configures it.

Try it — paste to your agent

Add a Chirp critic to my Grasshopper definition that reviews each layout option
against my brief and flags the weak ones.

Related

• Rook in Grasshopper