Kaer has one agent — the operator — and behind it, a stack of frontier models. You never pick a model, compare providers, or tune settings. You describe an outcome, the operator routes the work automatically, and every turn is metered into credits so you can see exactly what it cost. This page explains how that routing works and what it means for you.
Automatic model routing
When you start a turn, the operator reads what you are asking for and chooses how to run it. Quick work — a lookup, a small edit, a short draft — takes a fast path, so simple things stay fast. Hard work — deep research, a complex build, multi-step reasoning — goes to deeper models with more room to think. You see the result, not the machinery.
Routing is not one decision per conversation. Within a single turn, different branches can take different paths — a cursor doing a mechanical rename does not need the depth of one restructuring your data model. Matching the model to the branch is a large part of why turns feel quick without cutting corners on the parts that are genuinely hard.
No model picking
There is no model dropdown in Kaer, and that is deliberate. Model names change monthly; the shape of your work does not. Kaer keeps the stack current behind the scenes and holds one contract steady: describe the outcome, get the best available handling for it.
If a piece of work deserves extra depth, say so in plain language. "Take your time on this" or "think it through carefully" steers the routing far better than a model name ever would. The same works in reverse — "a quick answer is fine" keeps a simple question on the fast path.
Metered per turn
Every turn is metered into credits based on the work it actually used. Fast paths cost little; deeper models cost more. You can see what each turn cost in your usage, and when credits run out the operator pauses and waits — nothing bills silently in the background. See Billing & Usage for how credits work on each plan.
Supported capabilities
Across the stack, the operator handles:
- Text — drafting, summarising, editing, and rewriting
- Research — live search, source analysis, and synthesis
- Code — writing, reviewing, debugging, and explaining code
- Reasoning — multi-step analysis, planning, and decision support
- Structured output — tables, JSON, formatted reports
Each capability draws on the models best suited to deliver it. The routing is invisible to you but tuned for quality.
AI steps in workflows
When you add an AI step to a workflow, you do not pick a model there either. Kaer reads what the step is doing and routes it automatically — a fast path for a quick classification, a deeper model for nuanced drafting or analysis. Your workflows stay quick where they can be and high-quality where it counts, with no tuning on your part.
You write the instruction; Kaer chooses and manages the model behind it. The same routing applies whether the step runs once or fires on a schedule a thousand times.
Where to go next
To see the turn this routing lives inside, read The Operator. For how search specifically works, see Search & Research, and for what turns cost, see Billing & Usage.