Skip to main content
Open an Instance and go to the AI tab. Agent nodes on the Graph use these settings. Secrets never go to the model. The Agent only sees that a key is connected.

Provider

Where this Instance gets models:
  • Riggery — curated catalog. Chat uses your Included AI usage pool.
  • Your API — OpenAI, OpenRouter, or OpenAI-compatible. Chat uses your key; Memory indexing still uses Included AI usage.
See remaining Included AI usage on Usage.

Model

Pick a model from the provider list, or type an id. Vision, tools, and thinking follow what the catalog knows about that model. Some Riggery catalog models are marked Pro. Each model shows its price per 1M tokens (in / out).

API key

For Your API, bind a Secret. Do not paste keys into Chat or Settings text. Riggery does not ask for an API key.

Context

How much of the conversation the model can see at once. Leave empty to use the model default, within the maximum shown on the field.

Generation

Temperature, Max output tokens, and Reasoning effort apply to every Agent that does not set its own values. Leave Temperature on Provider default. 0 always picks the most likely words. Max output tokens caps the reply. On models that think, thinking tokens count toward this cap. An empty field uses 8192. Reasoning effort is Minimal, Low, Medium, or High. It is available when Native reasoning (thinking) is on. Provider default leaves the level to the provider. Turning thinking off stops reasoning on the next run.