Providers & models
A provider is the AI service that does the actual thinking for your workers — and askTheodor lets you mix several, picking the right brain for each job.
What you’ll learn
- The kinds of providers you can connect.
- How to give each worker its own model.
- Why local models are free and private.
Connect a provider
- Open Settings → Providers (the tab Settings opens on by default).
- Connect at least one — paste an API key for a cloud provider (OpenAI, Anthropic, Gemini, and others), or point at a local runner (Ollama, LM Studio). There’s also a zero-setup Built-in model that always works.
- A few seconds after you paste a key, the app quietly health-checks it and shows ✓ Working — you don’t press anything.
Beyond the headline backends, the provider directory reaches far wider — on the order of 49 providers you can wire up.
A model per worker
Each worker has its own brain — set on the worker, not globally. So the same profession can be a frontier-model strategist or a cheap-model clerk depending on the model you give it. Mix freely: a top model for the thinker, a lean one for routine clerks.
Three ways people set this up
There’s no single right configuration — it depends on what you’re optimising for. Three that make sense:
Zero setup, zero cost, zero accounts. Use the built-in model and nothing else. Nothing leaves your machine, there’s no key to manage and no bill to watch. Output won’t match a frontier model, but for organising files, summarising your own documents and drafting internal notes it’s genuinely enough — and it’s the only configuration that works with the Wi-Fi off.
One cloud key, used sparingly. Connect a single cloud provider and give it only to the two or three workers doing your hardest thinking; leave everything else on local. This is the best value/quality trade for most people, and the cost stays legible because you know exactly which workers can spend.
Local for anything sensitive, cloud for the rest. Point the workers that touch contracts, customer data or anything under NDA at a local model, and let the rest use cloud. The split isn’t a policy you have to remember — it’s a per-worker setting, so the sensitive worker can’t send data outward even if you forget.
Recap
You connected a provider, learned that every worker carries its own model, and saw that local runners are free and fully private. That rounds out the Connect track: local MCP, hosted connectors, one-time OAuth sign-in, and the brains behind it all.
📖 Reference in the docs: Providers · Models
➡️ Next: Back to the curriculum →