Sakana AI

Sakana AI is available in RondoFlow as a transform node (sakana-ai), not as an Assistant provider. It runs a Sakana chat-completions request mid-workflow, then passes the result to the next step.

This is useful when you want a dedicated model call between Assistant steps, for example to rewrite, normalize, or classify an upstream output before routing it onward.

Model presets

The node drawer ships with these model presets:

Model id	Notes
`sakana-chat`	Default preset for new nodes
`sakana-mini`	Smaller/faster preset

The model field still accepts any custom model id.

Setup

Set your API key

Add this to your server environment:


SAKANA_API_KEY=

The generated .env.example already includes SAKANA_API_KEY, but it is optional and only needed when you run Sakana AI nodes.

Add the node

From the canvas palette, drag Sakana AI onto the workspace.

Configure the request

Set apiUrl, select a model preset (or enter a custom model id), write the prompt, and optionally set system prompt, temperature, max tokens, and output format (text or json).

If SAKANA_API_KEY is missing, the step fails fast with a clear error (SAKANA_API_KEY is not set...).

How it runs

Interpolates {{input}}/{{output}} into the node prompt.
Calls Sakana’s chat-completions endpoint (https://api.sakana.ai/v1/chat/completions by default).
Returns either extracted completion text (outputFormat: text) or the full JSON response (outputFormat: json).

Sakana AI nodes run in both interactive and server-side execution paths because both compilers include sakana-ai as an executable step type.

Providers overview Data Nodes Workflows