{% extends "admin/base.html" %} {% block title %}Models Manager{% endblock %} {% block header_title %}Models Manager & Auto-Routing{% endblock %} {% block content %}
Ground model metadata in industry standard scorecards
The Models Manager defines the "Intelligence DNA" for the virtual auto model.
Instead of managing server hardware here (which has moved to **Workflows**), this page focuses on tagging models so the Gateway knows how to route requests based on prompt content.
Required for vision tasks. If a user attaches an image, the 'auto' router filters for these models only.
Enable for models supporting internal reasoning (DeepSeek R1, etc). Prevents 'think' parameter stripping.
Tagged as programming experts. Used when the router detects code keywords (e.g. 'def', 'class', 'import').
Small, optimized models. Used when 'fast_model' option is set in the request for low-latency tasks.
Complex logic models. Triggered by keywords like 'solve', 'math', 'why', or 'step by step'.
Token limit. The router checks if the current chat history fits before selecting the model.