Platform
Model Routes
Choose primary and fallback models for each processing capability and AI tier.
What this menu does
Route product AI workloads through controlled model chains with timeouts and retries.
Where to find it
Platform Admin navigation: LLM Management → Model Routes.
When to use it
- Review image and text processing routes.
- Edit primary/fallback models.
- Set timeout and retry limits.
- Test and enable/disable a route.
Before you begin
- Sign in to Platform Admin.
- The account must be allowed to view this menu.
Page regions
- 01Status filter
- 02Capability and AI-tier table
- 03Primary/fallback columns
- 04Timeout/retry columns
- 05Edit dialog
- 06Route test and status actions
Supported operations
Edit a model route
- 1
Select Edit on the route.
- 2
Choose the primary model and ordered fallback models.
- 3
Enter a positive timeout in seconds and a non-negative retry count.
- 4
Enter a change reason and save.
The route uses the saved model chain for later requests.
Test a route
- 1
Select Test on the route.
- 2
Wait for all configured attempts to finish.
- 3
Review latency, selected model, whether fallback was used, and attempted order.
- 4
Repair disabled or failing models before another test.
The test reports the selected model and attempted chain without changing business records.
Business rules and fields
- A primary model is required.
- Timeout is a positive integer in seconds; retries are non-negative integers.
- Fallback order matters.
- A successful fallback test indicates degraded primary availability, not a clean primary result.
States you may see
What happens next
- Each capability and AI tier has an explicit, testable model-selection policy.
Common problems
The menu or action is missing.
The account does not have the required platform access or the item is not available in the current state.
Ask a platform administrator to verify the account role and the record status.
A save or load action fails.
The session may have expired, validation may have failed, or the service is temporarily unavailable.
Keep the shown request identifier if present, refresh once, correct highlighted fields, and retry. Escalate with the request identifier if it repeats.