Pick the Right Brain for Every Task: The Effort × Model Slider

Stop sending “format this JSON” to your most expensive model. TinkerClaw’s effort × model slider lets you pick the right brain for every task: a cheap fast model for breadth, a frontier model for the hard reasoning, and an Auto mode that makes the call for you, turn by turn. Part of the open Building Jarvis series.

Two dials, not one

Model and effort are different questions. Model asks whether it can do this at all: task difficulty. Effort asks how much thinking this instance needs: task depth. A simple lookup on a strong model is waste. A hard problem on a weak model is worse, because it burns tokens circling a wall it cannot climb.

Each tile is a model, coloured by provider. Pin one and you set its effort yourself, from Minimal to Max. Put the model on Auto and the effort row turns into a bias dial, from budget to smart, and the router picks the effort for each turn.

Auto mode

On Auto, the router (I call it Thalamus) reads each turn and picks for you, leaning toward the bias you set and the kind of task in front of it. Breadth work goes to a fast cheap model, routine work sits in the middle, hard single-shot reasoning escalates. Its card in the same panel shows which model is answering, at what effort, and how big the job is.

The rule it follows saves the most money: escalate the model before the effort. A fix that keeps failing does not need more thinking on the same model. It needs a stronger model.

Why it matters

Strong-model-everything is how agent bills explode. The slider turns model spend into a choice instead of a default, and Auto makes the cheap-when-possible, strong-when-needed call thousands of times a day so you don’t have to. To see whether it chose well, the EEG draws every call it made and what each cost.

The deeper version

The theory of which work deserves which resources is in two papers of this series: the SALIENCE pyramid, which ranks attention before it is spent, and THALAMUS, the router itself.

TinkerClaw is open source. The whole system, from the brain-region modules to the recipes, the router and the EEG, lives at github.com/globalcaos/tinkerclaw. This post is part of the open Building Jarvis series at The Tinker Zone.