Model
The agent-model config node identifies one model available through the
agent’s selected connection. It can be shared by multiple agents.
Configure models in Node-RED: The current Agent Studio beta editor uses model fields that do not map reliably to the deployed
agent-modelschema. Set and verify Model ID, Temperature, Max Tokens, and Top P on the Node-RED configuration node.
Properties
Section titled “Properties”| Property | Range or default | Purpose |
|---|---|---|
| Name | Optional | Friendly label shown in configuration selectors. |
| Model ID | Required | Exact provider model identifier. |
| Temperature | 0 to 2, default 1 | Controls sampling randomness when supported by the provider. |
| Max Tokens | Positive number or empty | Limits generated tokens per model response. Empty uses the provider default. |
| Top P | 0 to 1 or empty | Nucleus-sampling setting when supported. Empty uses the provider default. |
Model IDs are passed to the configured provider. Use the exact identifier expected by that OpenAI-compatible endpoint.
Each Model config node contains one Model ID. The Agent node decides which config node fills each model role:
| Role | Use |
|---|---|
| Model | Direct reasoning and structured action output. |
| Think Model | Plain-text reasoning in Thinker mode. |
| Schema Model | Converts Thinker reasoning into structured runtime actions. |
When no Schema Model is selected, Thinker reuses the Think Model for extraction but still makes a separate second call.
Choosing Settings
Section titled “Choosing Settings”Lower temperature often improves repeatability and structured extraction. Higher values can increase variation but may make tool parameters or action JSON less consistent.
Top P is another sampling control. Unless a provider recommends otherwise, avoid aggressively changing both Temperature and Top P at the same time.
Max Tokens must be large enough for the complete response. Schema extraction can require more output space when the agent has many tools or complex parameters.
Compatibility
Section titled “Compatibility”Direct and the Thinker Schema Model require an endpoint and model that can produce the requested JSON object response. Provider support varies even among APIs described as OpenAI-compatible.
If normal chat works but agent actions fail, verify structured-output support, the model ID, and the response-token limit.
Agent Studio Editing
Section titled “Agent Studio Editing”The current Studio side panel labels model fields as Primary and Secondary and does not expose Max Tokens or Top P. Those fields should not be treated as the authoritative deployed configuration. Open the full Node-RED editor to create or change Model config nodes and to select the Agent node’s Think Model and Schema Model roles.