Smart Chat (Auto-Select Model)
AI-generatedSummary
Selects the optimal language model for a chat task automatically based on the specified task category, budget, domain, and filter criteria, then sends the provided messages to this selected model to receive a chat completion response.
Inputs
- Task Category (required) — Specifies the type of task the model should perform (e.g., Translation, Coding, Chat). This guides model selection by relevant benchmarks and capabilities.
- Budget (required) — Determines how much cost is acceptable, influencing model choice to balance price and quality (options: cheap, balanced, premium).
- Domain — Optional domain string to override routing policies for model selection (e.g., "bitcoin"). Leave empty to use default policies.
- Model Override — Optionally bypass automatic selection by specifying a particular model ID to use.
- Messages (required) — A list of messages forming the chat conversation, each with an author role (system, user, assistant) and content, sent as input to the selected model.
- Filters — Additional constraints to limit model candidates by minimum context length, JSON output support, vision capability, max cost per 1k tokens, and provider inclusion/exclusion lists.
- Options — Parameters that control chat generation behavior such as temperature, max tokens, JSON output enforcement, top P sampling, frequency and presence penalties.
Output shape
a single JSON object containing the chat completion result, including the generated content, model used, finish reason, usage metrics, unique ID, and input; augmented with the selected model's metadata (ID, display name, scoring details, pricing, reasoning) and routing metadata if applicable.
The operation uses a remote model catalog and routing policy fetched via configured credentials to select or validate models. If a manual model override is provided, selection and scoring logic are bypassed. If no suitable model is found, an error is raised. The output includes detailed information about the selected model and selection rationale. Errors from the OpenRouter API or selection steps are propagated unless 'Continue On Fail' is enabled.