Model Redirect

Automatically generates "Standard Model → Actual Model" mappings for New API channels, allowing you to call unified model names (e.g., gpt-4o) downstream, with the system automatically routing to the actual models provided by various upstream gateways.

Feature Overview

• Unified Model Invocation Entrypoint
  • You can uniformly use a set of "standard model names" (e.g., gpt-4o, claude-4.5-haiku) in New API.
  • Model Redirect will map them to the actual available model names for each channel.
• Automatic Mapping Generation
  • Automatically calculates the most suitable correspondences based on the models list reported by channels and a set of "standard model lists."
  • Supports incremental merging by channel, preserving your manually configured mappings.
• Use in Conjunction with Model Synchronization and Channel Management
  • Typically used with New API Model List Synchronization and New API Channel Management:
    • Model Synchronization is responsible for pulling the upstream available model list into New API.
    • Model Redirect is responsible for aligning these models to a set of "standard models."

Prerequisites

• New API Integration Settings have been configured in the plugin:
  • Fill in Admin URL, Admin Token, and User ID.
• At least one valid channel exists in New API, and the channel's models field contains a list of available models.
• A model synchronization has been completed (recommended) to ensure the model list within the channel is up-to-date.

Accessing Settings

1. Open the plugin → Navigate to the Settings page.
2. Switch to the "New API" tab.
3. Locate the "Model Redirect" area (id="model-redirect") at the bottom of the page.
4. You will see:
   • An enable toggle
   • A multi-select list for standard models
   • A "Regenerate Mappings Now" button

Key Configuration Items

• Enable Model Redirect
  • When enabled, the Model Redirect algorithm will be invoked when you manually click "Regenerate" or during specific synchronization processes.
• Standard Model List
  • Built-in standard models from common vendors, for example:
    • OpenAI Series: gpt-4o, gpt-4o-mini, o3, o3-mini, etc.
    • Mainstream models from Anthropic / Google / Mistral / DeepSeek / Qwen, etc.
  • You can add or remove as needed, or manually enter custom standard model names.
• Regenerate Mappings Now
  • After clicking the button, the plugin will:
    • Fetch all channel lists (filtering out manually/automatically disabled channels).
    • Calculate the "Standard Model → Actual Model" mapping table for each channel.
    • Merge the results with the existing model_mapping and write them back (new keys overwrite old values).
• Clear Model Redirect Mappings (Dangerous Operation)
  • Clicking this will first open a Preview/Selection modal, loading the channel list for the current site:
    • Select All channels by default;
    • Supports Select All / Deselect All quick actions;
    • Supports individually selecting channels to clear.
  • After clicking "Continue" in the preview modal, a secondary confirmation (unreachable by accident) will appear:
    • Upon confirmation, model_mapping = "{}" will be written to the selected channels (clearing mappings).
    • Channels not selected will not be modified.
    • This operation will clear any custom mappings you may have manually maintained and cannot be undone.
  • If partial failures occur:
    • The plugin will continue to attempt clearing other channels;
    • The UI will provide a summary of successes/failures and display failure details in the modal for you to retry or handle manually.

How It Works (Briefly)

1. Read the Model Redirect configuration from user preferences:
   • If not enabled, return directly and prompt "Feature is disabled."
2. Calculate the standard model set:
   • If you haven't customized it, use the built-in complete set of standard models.
3. Obtain all channel lists through the New API management interface.
4. For each channel:
   • Parse the channel's models field into a list of actual models.
   • Perform multi-stage matching for each standard model:
     • If the channel itself contains a model with the same name, skip (no redirection needed).
     • Otherwise, find the closest candidate actual model based on the unified "Model Normalization Rule" (renameModel).
     • Version Guard: Only match models of the same version; do not downgrade/upgrade across versions (e.g., claude-4.5-sonnet will not be mapped to claude-3.5-sonnet; if no same-version candidate is found, skip this standard model).
     • Select an unused actual model from the candidate set and write the standardModel -> actualModel mapping.
   • Merge the new mappings with the channel's existing model_mapping JSON and write them back.
5. Summarize the number of successfully updated channels and reasons for failure, displayed as a toast in the UI.

Typical Use Cases

• Unifying OpenAI Compatible Model Names
  • Different upstream providers may use slightly different model naming conventions (e.g., with prefixes/suffixes).
  • Through Model Redirect, you can consistently use unified names in downstream applications, such as gpt-4o, with New API routing to the actual model based on the mapping.
• Coordinating with Channel Priority
  • Set weights/priorities for different channels in New API, and then combine with unified standard model names to achieve multi-upstream disaster recovery and traffic distribution.

Frequently Asked Questions

• Prompting for Missing New API Configuration
  • Please first fill in the Admin URL, Token, and User ID in "Basic Settings → New API Integration Settings" and save.
• Prompting that the Feature is Not Enabled
  • Confirm that the "Enable Model Redirect" toggle has been turned on in the "Model Redirect" area.
• Some Standard Models Did Not Generate Mappings
  • This indicates that no matching actual model was found in the corresponding channel, or all candidates were already occupied by other standard models.
  • It's also possible that the channel has no same-version candidate models: the plugin will not generate mappings by downgrading/upgrading across versions.
  • You can check if the channel's models field is complete, or adjust the standard model list.
• Will Existing Manual Configurations Be Overwritten?
  • For the same standardModel, newly generated mappings will overwrite old values;
  • Other keys added manually but not overwritten will remain in model_mapping.

Usage Recommendations

• It is recommended to first validate the generation results on a test environment or a small number of channels before performing batch redirection on all channels.
• If you primarily maintain your model list through New API Model Synchronization, it is recommended to click "Regenerate Mappings" after completing the synchronization.

Related Documentation

• New API Model List Synchronization
• New API Channel Management
• Quick Export Site Configuration