Smart Model Router
Smart Model Router is an option in the model selector. When it's the active model, AI CTRL doesn't send every message to a single model you've chosen. Instead, for each message you send, it reads the Description field of your available custom models and picks the one that best matches what you're asking.
It only chooses among custom models: any custom model you've created, plus any public custom model shared with you by someone else in your organization. It doesn't route to the built-in models listed on the Available Models page.
Smart Model Router is useful when you're not sure which of your custom models fits a given request. If you already know which model you want, select it directly from the model list instead of routing through Smart Model Router: you get that model's output without depending on how closely its description matches the wording of your message.
How the match happens
The router re-evaluates on every message, not once per conversation. It compares your message against the Description field of each custom model available to you and sends the message to the closest match.
Because matching happens per message, a single conversation can move between different custom models if your questions change topics partway through.
Writing descriptions that route well
The Description field is the only thing the router reads to decide. A vague or overlapping description produces inconsistent routing, not a failure, just an unpredictable one.
| Description | Routing outcome |
|---|---|
| "Helps with writing" | Matches almost anything, so it competes with every other model you own and wins unpredictably. |
| "Drafts and edits customer support replies, focused on a friendly tone for billing and account questions" | Matches a narrow set of prompts, so it's selected consistently and only for that kind of request. |
| "Reviews vendor contracts for indemnification and liability language" | Specific to a task and a domain, so it doesn't compete with a general-purpose model you've also created. |
Two practical rules follow from this:
- Describe the task and domain, not the model's personality or tone. "Reviews contracts for X" routes better than "a sharp, no-nonsense assistant."
- Keep descriptions distinct from each other. If two of your custom models have descriptions broad enough to both plausibly match the same prompt, routing between them becomes inconsistent.
Seeing which model answered
During generation, the thinking indicator shows the name of the model the router selected for that message.
When nothing matches
If none of your available custom models match a message, or you have no custom models available to you, AI CTRL falls back to a task model that your administrator has configured for this purpose. You don't control this fallback model and can't change it yourself.
If that fallback model has been deprecated or is a private model you don't have access to, you'll see an error instead of a response. This isn't something you can fix from your side: contact your administrator, since the fallback model is an organization-level setting.