AI & Machine Learning

askTheodor — a local-first AI workforce for companies of one

A desktop app that staffs a one-person company with AI workers for sales, ops, research and content. Theodor, the chief of staff, splits the work, hands it to specialists and reports back; nothing leaves the machine without your approval. It started in 2024 as a web chat with eight AI agents, and relaunched on 17 July 2026 as a native app for macOS and Windows.

Tauri 2 Rust Tokio Axum SQLite (sqlx) React TypeScript Vite Zustand Three.js MCP Ollama Anthropic OpenAI Gemini RAG
askTheodor — a local-first AI workforce for companies of one

Key Features

  • AI workers from 95 professions, each with its own persona, remit, tool allowlist, voice and long-term memory
  • Theodor, the chief of staff, turns a big request into a plan (a DAG of tasks), routes each task to the right worker and summarises the result
  • Agents prepare, humans approve: an approvals queue, a risk gate, a kill switch and a full audit log
  • Local-first: prompts, files, embeddings and credentials stay on your disk; bring your own keys for Anthropic, OpenAI or Gemini, or run offline with Ollama
  • Companies (per client or project), each with its own Library and RAG index; workers cite their sources
  • Overnight work with routines and a daily digest of what shipped
  • Real ops tools: persistent processes, SSH/SCP deploys, spreadsheet and PDF parsing, code search
  • Fleet: deploy headless askTheodor instances to your own servers over SSH, running 24/7 and reporting back
  • Voice chat and read-aloud with a voice per worker; a kiosk mode turns it into a multilingual touch-screen receptionist
  • 116 task templates plus guided playbooks for launching a business

Challenges Overcome

  • Rebuilding a cloud chat product as a native desktop app without losing the Theodor persona people knew from v1
  • Keeping everything local-first while still letting users pick any cloud model, or none at all
  • Making delegation trustworthy: every hop between workers is logged, budgeted and can be paused
  • Scheduling plans as dependency graphs, with retries, that keep running while the laptop owner sleeps
  • An approval gate strict enough that nothing goes out without a yes, but light enough not to slow the day down
  • Earlier on: real-time streaming chat across several agents, token billing, and Stripe + PayPal payments in four languages (v1)

Results & Impact

Launched 17 July 2026 as a desktop app for macOS and Windows; v0.9.279 on the website as of September 2026
Simple pricing: 7-day free trial with no card, then one plan with every feature at €12/month or €120/year; add-ons for kiosks and remote instances
Rust workspace of 10 crates (agent core, memory, MCP client and server, LLM providers, licensing and more) behind a React front end
A done-for-you setup and training service for teams that don't have time to evaluate it
History, v1 (2024–2025): a web chat platform with 8 specialised agents (Theodor, CodeCraft, TaskMaster, WordSmith, SalesGuru, MarketMaster, EduMentor, BizPro) in English, German, Greek and Turkish, built with React, Vite, Supabase, the OpenAI API, Stripe and PayPal, on Free, €30 and €60 plans
History, v2 (2025): a rebuild on Next.js, AWS Amplify and LangChain, with the first MCP experiments that led to the current design
Theodor himself is the thread through every version: the chief of staff today is the first agent from 2024

"askTheodor began as a chatbot with a few clever personas. It became the team I wished I had while running a company of one: workers that remember, delegate and report back, on my own machine, and never send anything without my yes."

Theodoros Dimitriou — Creator of askTheodor