Routing Groups & Fallbacks

V2 console

This card lives in Policy Engine > LLM Gateway at web.quilr.ai/policy. Edits join the shared draft and take effect once you review and publish a revision.

Decide where a request is served: an ordered fallback chain of provider credentials, a weighted split across models, or word-count thresholds that classify prompts as Low, Medium or High complexity.

Order of decision

Allowed Models

Filters what may be used

Route to

Highest-priority route

First enabled target

Routing group

Highest-priority group

Weighted split

Provider default

App's own provider

QuilrAI

A route beats a routing group for the same traffic. If neither matches, the application's provider default serves the request, and the request keeps the model it asked for.

Sections

Section	What it does	Applies on	Empty state
Route to	An ordered list of provider credential and model pairs. The first with an enabled credential serves the request, whatever model the client asked for. Add route.	bedrock, chat, responses, vertex	Requests keep the model they asked for.
Routing groups	A weighted split across models. Weights always total 100. New group.	bedrock, chat, realtime, responses, vertex	No split.
Complexity thresholds	Word counts that classify a prompt Low, Medium or High for the Prompt complexity scope chip. Set thresholds.	bedrock, chat, responses, vertex	6 words or fewer are Low, 7 to 10 are Medium, longer are High.

Configuration settings

All three buttons open one New Routing Groups & Fallbacks configuration dialog. Send to picks the mode.

Setting	Options	Default	Notes
Applies to	Everyone, People, Smart group, Application, App tag, Requested model, Provider, API surface, Environment, Prompt complexity, Prompt text, Tool, Source network, Except...	Everyone	Or pick an application directly. Thresholds cannot be scoped by Prompt complexity.
Send to	Ordered targets, Weighted split, Complexity thresholds	Depends on the button
Severity	Not set to Very critical	Not set	Reported only.

Ordered targets (Route to)

Setting	Options	Notes
Targets	Provider credential + model, one row per target	Tried in order. Drag the handle or use the arrow keys to reorder. Add target for more fallbacks.

Weighted split (Routing groups)

Setting	Options	Default	Notes
Name	Text, required	-	A policy-only alias. It never affects a live routing group with the same name. Converted app groups are scoped with Requested model set to the group name.
Kind	Chat completion, Anthropic messages, Vertex AI, Responses, Realtime, Bedrock	Chat completion	The API shape the group serves.
Mode	Split by requests, Split by tokens	Split by requests
Models	Add model per provider credential + model, each with a weight	-	Weights must total 100. Balance evenly splits them equally.

Complexity thresholds

Setting	Default	Notes
Low up to	6 words
Medium up to	10 words	Anything longer is High.

Examples

support_failoverrequest

runs on request·priority 500

WhenApplicationisSupport Copilot

Then

Route to→2 targets

1 openai_primary / gpt-4.12 azureopenai_primary / gpt-4.1

OpenAI serves every Support Copilot request. If that credential is disabled, Azure OpenAI takes over.

cheap_short_promptsrequest

runs on request·priority 600

WhenPrompt complexityislow

Then

Route to→openai_primary / gpt-4.1-mini

Short prompts go to a cheaper model. Combine with Complexity thresholds to change what counts as short.

Scoping and precedence

• Highest priority wins where scopes overlap. New configurations start at 500.
• Rejected models on Allowed Models still win over a route or group target.

Legacy app setting

Request Routing in the app's settings.