Pick a brain per worker

A worker’s brain is the model and provider it thinks with — and you set it per worker, so each one runs on the right brain for its job.

What you’ll learn

• How to set a worker’s model and provider.
• How to match brain to role (and budget).
• Why local models are worth it for routine work.

Brain = model + provider

On a worker’s profile, ✎ Edit opens its character sheet; the brain is the model (the specific AI doing the thinking) plus the provider that supplies it. askTheodor connects to 49 providers — OpenAI, Anthropic, Gemini, local runners, and many more — and different workers can use different ones.

Connect one to start

You only need one provider connected to begin. Add more later so you can mix brains across your workforce.

Match the brain to the role

The same profession can run on very different brains. A useful rule of thumb:

• Frontier models for strategists — the workers doing planning, judgment, and high-stakes writing earn a top-tier brain.
• Cheap or local models for clerks — routine, high-volume, low-risk tasks don’t need the most expensive model.

Local = free and private

Point a worker at a local model and its thinking is free (no per-token cost) and private (the data never leaves your machine). Ideal for clerks churning through routine work, or anything sensitive.

What this lets you build

Per-worker brains let you spend money where it changes the answer, and nothing where it doesn’t. Three arrangements that work:

A cheap clerk feeding an expensive strategist. Put a local model behind the worker that sorts, extracts and summarises raw material, and a frontier model behind the worker that decides what to do about it. The bulk work costs nothing and never leaves your machine; you pay only for the thinking that genuinely needs judgement.

A private worker for the sensitive pile. Point one worker at a local model and give it the material you’d never paste into a cloud chat — contracts, customer records, anything under NDA. Its answers will usually be less polished than a frontier model’s, and that’s the real trade: you’re buying privacy, not peak reasoning.

A workforce that survives a provider problem. Because the brain is set per worker and there are 49 providers, an outage or a price rise is an edit on a profile rather than a rebuild. Switch that worker to another provider — or to local — and its job, memory and tools carry on unchanged.

Recap

Each worker gets its own brain from 49 providers — frontier for strategists, cheap or local for clerks, with local models free and private. Next, decide which tools each worker is even allowed to hold.

📖 Reference in the docs: Models · Providers

➡️ Next: Tool allowlists & least privilege