Model Advisor

The Model Advisor is Yukkai's local task classification system that analyzes each message and recommends or routes to the best model — powering Fusion and Frontier's automatic routing.

The Model Advisor is Yukkai's local task classification system. When you choose to use Yukkai Fusion or Frontier, the Model Advisor reads your message, analyzes it, and routes the request to the LLM best suited to handle it — always keeping the cost/quality ratio in mind.

What is the Model Advisor?

The Model Advisor is a local system on your Mac (part of the RoutingEngine) that:

Reads your message — analyzes the content, structure, and intent.
Classifies the task — determines whether it's a chat question, code problem, reasoning task, vision request, long-context analysis, etc.
Routes to the best model — selects the most appropriate model from the active pool (Fusion or Frontier) based on the classification.

All of this happens locally on your Mac — the classification doesn't send your message anywhere before routing it.

How It Works

For Yukkai Fusion and Frontier

When Fusion or Frontier is active, the Model Advisor is fully automatic:

You send a message.
The Model Advisor classifies the task locally.
It selects the best underlying model from the Fusion or Frontier pool.
The request is sent to that model.
You get the response — you never see which model was chosen.

This happens every single message — the underlying model can change from one message to the next depending on what you asked.

Task Classification Categories

The Model Advisor evaluates tasks across multiple dimensions:

Category	Examples	Typical Model Choice
Chat	Quick questions, casual conversation	Fast, light model
Code	Bug fixes, refactoring, code generation	Code-specialized model
Reasoning	Logic puzzles, math, strategic planning	Reasoning-optimized model
Vision	Image analysis, OCR, visual QA	Vision-capable model
Long-context	Analyzing large documents, multi-file code review	Long-context model
Creative writing	Stories, poems, marketing copy	Creative-optimized model

The exact classification logic and model selection criteria are proprietary — Yukkai doesn't disclose the scoring/weights involved.

Model Advisor vs. Fusion/Frontier

Don't confuse these related but distinct concepts:

	Model Advisor	Yukkai Fusion/Frontier
What it is	The local classification and routing system	The selectable meta-model identities
What it does	Analyzes your message and picks the best model	The pool of models the Advisor routes to
Visibility	Runs invisibly in the background	You select "Fusion" or "Frontier" in the model picker
User control	None (fully automatic)	You choose which tier to use (Fusion vs Frontier)

Think of it this way:

Fusion/Frontier = the brand/pool of models.
Model Advisor = the engine that routes within that pool.

The Separate Model Advisor Feature

There's also a separate model-advisor feature that helps users manually choose among regular (non-meta) models. This is distinct from the Fusion/Frontier auto-routing system.

The manual model advisor:

Suggests which specific model might be best for your task.
You make the final choice from the recommendations.
Works with personal OpenRouter models, not Fusion/Frontier.

This wiki article focuses on the automatic routing system used by Fusion and Frontier.

Privacy and Performance

Local Processing

The Model Advisor runs entirely on your Mac — task classification happens locally before the request is sent to any cloud model. This means:

✅ Your message isn't sent to a separate classification service.
✅ No extra network round-trip for routing decisions.
✅ Fast routing (no added latency).

What Gets Sent to the Cloud

Only your final routed request goes to the selected model:

The Model Advisor picks the model.
Your message is sent to that model via OpenRouter.
The response comes back.

The classification itself never leaves your Mac.

Why Automatic Routing?

The Model Advisor exists to solve a real problem: model choice is hard.

With 300+ models available, each with different strengths, how do you know which one to use for a given task? Most users don't want to think about:

Whether this task needs a code-specialized model or a general one.
Whether the extra capability of a frontier model is worth the cost.
Whether a fast, cheap model is sufficient or a slower, smarter one is needed.

The Model Advisor handles this for you — you just use Fusion or Frontier, and it picks the right model every time.

When the Model Advisor Matters

The Model Advisor is most valuable when:

You have varied tasks — some chat, some code, some research, some creative writing. The Advisor routes each to the best model.
You want to optimize cost — the Advisor keeps the cost/quality ratio in mind, so you don't overpay for simple tasks.
You don't want to think about models — you just want good results without managing model selection manually.

Limitations

No User Override

When Fusion or Frontier is active, you cannot override the Model Advisor's choice. If you want to force a specific model, you need to:

Switch to a personal OpenRouter key.
Select that specific model from the catalog.

No Visibility

You cannot see which model the Advisor picked for a given message. This is intentional — Fusion and Frontier are presented as unified Yukkai models, and the underlying model selection is proprietary.

No Explanation

The Model Advisor doesn't explain its routing decisions. If you ask why a particular model was chosen, Yukkai will say only: "Yukkai automatically selects the best available model for the task."

Examples

User asks about routing

User: "How does Yukkai decide which model to use?"

Yukkai (as Fusion): "Yukkai uses a local Model Advisor system that reads your message, analyzes the task, and routes the request to the best model for the job — always keeping the cost/quality ratio in mind. This happens automatically on every message."

User asks which model was picked

User: "Which model did you use for that last response?"

Yukkai (as Fusion): "I can't share that — Yukkai Fusion keeps the underlying model selection internal. What I can tell you is it's chosen automatically based on what your request needs."

User wants to force a specific model

User: "I want to use Claude for this task, not whatever Fusion picks."

Yukkai: "To use a specific model like Claude, you'll need to switch from Yukkai Fusion to a personal OpenRouter key and select Claude from the model catalog. Fusion doesn't allow overriding the automatic routing."

Good to Know

The Model Advisor is local — classification happens on your Mac, not in the cloud.
Routing is per-message — the underlying model can change from one message to the next.
You can't see or override the choice — that's the point of Fusion/Frontier.
The exact routing logic is proprietary — Yukkai doesn't disclose the scoring/weights.

Learn More

Yukkai Fusion — Everyday model routing powered by the Model Advisor.
Yukkai Frontier — Premium model routing for complex tasks.
Identity Lock Rule — Why the underlying model is never revealed.
Model Selector — Overview of all model options.
Personal OpenRouter Key — Manual model selection.