Autark is a free desktop app that runs AI agents as a team

Autark, an early desktop app whose source is published as the OpenAutark project by Austria's Teydra Lab, lets users run AI agents as a standing team with roles and ongoing tasks on their own computer, using model access they already have.

Illustration of a desktop computer on a home office desk, its screen showing a manager card routing work to several agent cards beside a task list. Image: AgenticFeed illustration

Autark, an early desktop app from Teydra Lab in Austria, organizes AI agents as a standing team, according to its website. OpenAutark is the open-source project and its GitHub repository, teydra/OpenAutark, Teydra Lab told AgenticFeed. A user hands work to a manager agent that passes it to specialist agents and reports back, the site says.

Stanislaus Hees created and leads the project, Teydra Lab said. AI coding agents write the code under his direction, among them Claude Code and Codex as well as Autark's own agent team, according to the company.

Teydra Lab answered AgenticFeed's questions on Nov. 17 through Paula, an AI colleague on Hees's team who handles press mail, according to the company.

How the team works

Autark's agents and their memory live in a folder on one PC, while model requests still go to the cloud providers a user connects, according to the site. A user types or dictates requests into one inbox. A manager agent can split a request into tasks and assign each to a specialist agent or to the user's own to-do list. Agents can wake on schedules, timers, messages and events while the PC is on, and their steps are written to a run journal the user can read, the site says.

The site also says agents have full access to the PC from the start. The manager first asks how far it may go, which the site calls "a question, not a barrier." In a Nov. 9 partner interview published by AI Business, Teydra Lab said the app should not be treated as a sandbox or as a guaranteed approval gate.

Bring your own model

The download is free and needs no account, and users can connect a ChatGPT plan through OpenAI Codex, a paid Claude plan through Claude Code, an API key or a local model server such as Ollama, according to the site's FAQ. Everything an agent works with goes to the connected provider, including private files it opens, the FAQ says.

All live agent runs the company can see use Claude Code or Codex on subscription plans, Teydra Lab said. The paths for API keys, local models, Gemini CLI, Copilot CLI and Kimi Code are built and tested but have not yet had a live run, according to the company. Teydra Lab also told AI Business it had not established that every aspect of its Claude integration meets Anthropic's conditions.

The installers are still unsigned, so Windows SmartScreen may warn before installing, according to the repository's README. Teydra Lab said a code-signing certificate is on order.

What the repository shows

The code is under the Apache License 2.0, with documentation and the skill library under CC BY 4.0, according to the repository. Teydra Lab said the desktop app's full source is public, including the agent runtime in apps/desktop, along with the Android app, the website and the meeting server, which is in the repository's services/remote-relay folder. The repository is a published snapshot that leaves out private history and internal notes, along with the generator for its model ratings, which is why its history starts on Nov. 6 and its public commits come from the teydra account, the company said.

As of Nov. 17, the repository had 30 stars and 4 forks, and its main branch had 24 commits, figures Teydra Lab confirmed. The lab has published seven desktop releases, from version 0.3.3 on Nov. 6 to 0.3.9 on Nov. 16, plus separate Android builds.

Teydra Lab, the sole proprietorship of Meral Anna Hees, is OpenAutark's legal publisher and media owner, according to the project's imprint.

Teydra Lab's Nov. 7 fact sheet gives two examples of the app at work, the company said. In one, its agents found a press inquiry Hees had missed. In the other, they checked the print file for his book "KI ins eigene Produkt," which is available free in German on his website. The company presents these as its own examples, not measured results.

Teydra Lab won a startup award at the 16th Austrian Innovation Forum in Vienna for its chatbot builder for company websites, not for OpenAutark, Trending Topics reported on Oct. 8, and Teydra Lab confirmed.

Source: Autark, “Autark: your own AI company that works on its own”

Additional sources: Autark, “Download Autark”; Autark, “Who builds Autark: Teydra Lab and Stanislaus Hees”; GitHub, “teydra/OpenAutark repository and README”; GitHub, “OpenAutark releases”; Autark, “Impressum (legal notice)”; Stanislaus Hees, Stanislaus Hees, “KI ins eigene Produkt”; AI Business, “OpenAutark: A Salzburg Lab's Free, Open-Source App Gives AI Agents Jobs, Memory and a Manager on Your Own Computer”; Trending Topics, Georg Haas, “Austrian Innovation Forum winners report, Oct. 8, 2026”

Get the AgenticFeed newsletter

A free, regular roundup of the agent stories that matter. No spam, and you can unsubscribe at any time.

Want to reach readers following AI agents? Advertise with AgenticFeed

Next event · 19 Nov, 2026

MCP Dev Summit Nairobi

The Linux Foundation’s Nov. 19 and 20 summit brings Model Context Protocol developers to Kenya, with keynotes from OpenAI, GitHub and the Gates Foundation.

The AgenticFeed newsletter

A free, regular roundup of the agent stories that matter. No spam.