What is askTheodor?
askTheodor is a desktop app that lets you build and run a team of AI workers. Think of it as a small company that lives on your computer: each worker has a name, a role, skills, and tools, and they do real work for you — research, writing, coding, design, planning, data entry, digesting your inbox, and preparing documents — while you stay firmly in control.
Who it’s for
Anyone who wants AI to actually do work, not just chat. Founders, freelancers, marketers, developers, analysts, operators — anyone who’d benefit from a tireless team that prepares everything for you to review.
The big ideas (in one minute)
- Local-first. Your workspace — companies, workers, documents, conversations — lives on your machine, in a local database. Nothing is uploaded to us. You bring your own AI provider keys.
- A workforce, not a chatbot. You don’t talk to one model; you build a roster of Workers (personas), organize them into Companies, and let Theodor — your AI chief-of-staff — coordinate them.
- Real tools. Workers can browse the web, read and write files, run shell commands (when you allow it), search a knowledge Library, make documents/PDFs, and more.
- Autonomy with guardrails. Workers can run Plans (multi-step projects) and Routines (scheduled jobs) on their own — but spending is capped, dangerous actions need approval, and nothing is emailed or posted without your sign-off. One switch pauses all of it.
- Batteries included. Sensible defaults out of the box: a starter worker, demo packs you can install, and Theodor ready to help. You can be productive in under an hour and only touch the deep settings when you want more control.
How it’s different from ChatGPT or a coding assistant
| A typical chatbot | askTheodor | |
|---|---|---|
| Where your data lives | The vendor’s cloud | Your machine (local-first) |
| What you interact with | One model | A team of named workers + an orchestrator |
| Memory | A single thread | Per-worker memory + a shared Library (RAG) |
| Doing work unattended | No | Plans & Routines run on a schedule, even headless |
| Safety rails | Generic | Budgets, per-tool limits, approval gates, a kill-switch |
| Cost | Subscription | Your own provider keys — you see every cent in Cost |
What you’ll need
- A computer (macOS, Windows, or Linux).
- At least one AI provider — a cloud key (OpenAI, Anthropic, Gemini, xAI, DeepSeek, …) or a local model runner (Ollama / LM Studio), or the tiny built-in model that ships in the app.
Ready? → Install askTheodor, then Your first hour.
In practice
Three shapes of work this structure handles well — all things a single chat window handles badly:
Work that arrives on a schedule rather than when you’re at the keyboard. A routine that digests overnight email and leaves you a briefing; one that checks a handful of sources each morning and writes what changed. The output is waiting when you sit down — you didn’t have to prompt for it.
Work with several steps that each need different skills. A competitor analysis is research, then synthesis, then a written document. As one long chat, you drive every handoff yourself. As a Plan, tasks go to the workers suited to them, the Reviewer checks the result, and you approve at the start and read at the end.
Work that has to be grounded in your own material. Your brand guide, your product spec, your past reports — put them in the Library and workers quote them instead of inventing plausible-sounding substitutes. That’s the difference between output you can send and output you have to fact-check line by line.
Where a chatbot is still the better tool: a one-off question, a quick rewrite, anything you’d finish in two exchanges. Configuring a worker for that is pure overhead. askTheodor earns its keep on work that repeats, has structure, or needs to happen while you’re elsewhere.
What “local-first” does and doesn’t mean. Your workspace lives in a database on your machine and nothing goes to us. But if you connect a cloud provider, your prompts do go to that provider — that’s how the model thinks. If nothing may leave the machine at all, run Ollama or the built-in model. See Security & privacy for the full picture.
🎓 Learn it hands-on: Install & first run
Terms in this page
- Local-first — the app stores your data on your own device first (here, a local SQLite database), rather than on a remote server. You keep ownership and can work offline.
- AI worker / persona — a configured AI character with a name, role, system instructions, allowed tools, and memory. Many workers make a workforce.
- Theodor — the special, built-in “chief-of-staff” worker that coordinates the others, drafts plans, and reports back to you.
- Tool — a capability an AI worker can invoke (web search, file read/write, shell, make a document, etc.) beyond just generating text.
- Plan / Routine — a multi-step project (Plan) or a scheduled/triggered job (Routine) that workers execute, often unattended.
- Library (RAG) — your curated knowledge base. RAG (Retrieval-Augmented Generation) means the app fetches relevant documents and feeds them to the worker so its answers are grounded in your material.
- Provider — the company or program that supplies the actual AI model (e.g., OpenAI, Anthropic, or a local runner like Ollama). You connect at least one.
- Approval gate — a checkpoint where the app pauses and waits for your OK before doing something irreversible or sensitive (like sending an email).