Available
Manager, teams and a secretary
Start with a manager. It builds the team you need (research, marketing, finance, a personal secretary) and hands work to the right person. Agents have names, avatars and their own instructions in a file.
Everything your AI company can do.
68 capabilities in 10 areas. 68 are in today’s public release. Nothing here is a mock-up.
How to read the labels. Available is in the current public release (0.3.2) for Windows, macOS and Linux; where a system differs, the text says so. We would rather undersell than show you something that is not there. What is not proven yet is on the access page.
01 5 capabilities
Your AI company
A manager, specialist agents and a private secretary, each with a name and a face. They split up the work among themselves.
Available
Delegation between agents
Agents pass tasks to each other, ask colleagues for help and report back when the work is done. You talk to one of them; the organisation does the routing.
Available
An agent can split itself
For work that runs well in parallel, an agent decides itself to split into parts, each working on its own, and brings the results back together.
Available
Bring agents you already have
Hand your manager Claude Code subagents, a Codex AGENTS.md, a CrewAI YAML file or a list of roles. A skill shows it how to bring them into the team.
Available
Humans and agents on one team
Invite colleagues by link or QR code. People are round, AI is square, and everyone works in the same chats and tasks. The data stays on your PC.
02 9 capabilities
Chats and work
Talk like you would in a messenger. Drop work into one inbox. Everything lands somewhere, nothing gets lost.
Available
Chats and group chats
Every agent has a direct chat; add groups and as many chats per agent as you like. You can keep typing while agents work; a message marked important steers work that is already running.
Available
Three message levels
Messages can be normal, important or quiet, in both directions, and every reply shows which model wrote it.
Available
Work: one inbox for every task
Type or dictate. With Auto your manager takes it, decides who does it and can split it into separate tasks. Or pick an agent yourself, or Me for your own to-do list.
Available
My day
Your own to-dos and calendar in one calm list, with learned time estimates, a morning plan and a short briefing of what happened.
Available
Team
Every agent and person, what they are working on and what they did. An agent’s page opens its instructions and its live browser.
Available
Calendar
Read any ICS feed, or connect Google Calendar both ways so agents can plan around your real day.
Available
Send later, or on a schedule
The message box has a schedule mode: a message or a routine goes out at the time you set and wakes whoever it is for.
Available
To-dos that become tools
A to-do can turn into the widget it needs: a recorder, a teleprompter, a timer, a form. Answers can contain tables, charts and diagrams.
Available
Tabs beside the chat
A tab strip like Chrome’s beside every chat, for websites and for the app itself: My day, Work, Activity, Notes, Media, Files, Screen and an agent’s browser. Back and forward, history, zoom per site, mute, Ctrl+T and friends.
03 5 capabilities
Proactive by design
Agents sleep until something happens. Then they wake up, work, plan their own follow-ups and go back to sleep.
Available
Wake-ups and routines
A message, a task, a timer, a schedule or an event wakes an agent. It decides what to do, does it, and can plan the next step itself.
Available
Keeps working in the background
Close the window and the company keeps running from the system tray, as long as your computer is on.
Available
Keep agents running
Optional and off by default: Start with Windows or Start at login in Settings. The app starts when you sign in and comes back after a crash. On Windows it can also start for the next due work after you quit.
Available
Results checked against the goal
A task can carry a done-condition. Its result is then checked: does the named file exist, does the link answer, and a decision model reads the result without the agent’s conversation. Without a done-condition or a decision model the result stays marked unchecked.
Available
Picks work up again
After a lost connection, a used-up plan or a restart, a retry comes back with growing gaps and tells the agent what happened and what changed meanwhile. Tested with kills and restarts on a test engine, not yet with real providers.
04 8 capabilities
The AI you already pay for
Autark is not another AI subscription. It runs on your existing plans through the official tools of each provider.
Available
ChatGPT and Claude plans
Agents work through Codex with your ChatGPT plan and through Claude Code with a paid Claude plan. Add as many accounts per provider as you like. The app can install either tool for you and test it with a real answer.
Available
Gemini CLI, Copilot CLI, Kimi Code
Built on the Agent Client Protocol: they sign in and run under the same rule as the other plans. Early: we have never seen them in a live run.
Available
The right model for each turn
Each turn gets an engine, an account, a model and an effort level. Live quota bars show how much of each plan is left, and you choose Best, Balanced or Cheaper.
Available
Use it before you lose it
Weekly limits expire. Among accounts of the same priority, the plan quota that resets soonest while some is left goes first.
Available
Switch engines without losing context
When an agent moves from Codex to Claude Code or back, its session is rebuilt from a shared context ledger. Nothing has to be explained twice.
Available
Lean context
Tools load when they are needed instead of filling every call, and long old output can be compacted, hidden or brought back.
Available
API keys and local models
Optional: OpenAI, Anthropic, Google, OpenRouter, DeepInfra, Groq, Mistral, xAI, DeepSeek, Kimi, Qwen, or local servers like Ollama, LM Studio, llama.cpp and vLLM. Early: built, no live run recorded yet.
Available
A model for each use
One row per use (work, decisions, voice, speech, images, video, music, sound effects, embeddings), each with its provider and its model. A name the app does not know can simply be typed in.
05 12 capabilities
An open system
No connector marketplace and no hidden settings. What the person or the AI may change is a plain file, so the AI can set up and extend the app by itself.
Available
A short start
Connect one AI, type your name and company, press Start. The manager gets to know you and your work, and first asks how far it may go on this PC.
Available
A browser per chat and per agent
Three places: the global browser where you work, a browser for each chat with tabs everyone in that chat shares, group chats too, and each agent’s own, which you can watch. Agents working in the background do not disturb your tabs.
Available
Providers through a file
Every API connection is one entry in settings.json: a name, an address, a protocol, a key as an ID. A provider the app has never heard of works if it speaks one of the protocols. Say the provider’s name and hand over a key; the AI can write the rest.
Available
Skills as open files
A skill is a folder with a SKILL.md, the format other agent tools use too. Each has a route: an agent’s own, its managers’, shared or built-in. The AI can read skills from folders of other tools when it chooses to, and writes new ones itself.
Available
Memory as open files
One Markdown file per memory, in folders for the company, a branch or a single agent. Open them, change them, version them. Any change is searchable at once.
Available
The AI builds its own connections
There is no WhatsApp button. Your agent sets it up: a small script next to the app, using the running app’s own interface. Telegram works the same way with a bot it creates. Other channels, such as Signal or Matrix, are the same kind of script: skills ship for WhatsApp and Telegram, and one that shows how for any other channel.
Available
Settings in one file
settings.json holds every setting with a comment above it. Change a value and save; it applies within a second. So you can just tell your AI what to change.
Available
Settings pages on that file
General, Models, Voice, Calendar, Secrets, Apps, Skills & memory, Devices, Feedback and About, with a search. Each page opens the file behind it.
Available
Files in a tab
A plain editor and a folder list beside the chat. Any path on your PC, Ctrl+S to save. Open settings.json, a skill, a memory or an agent’s instructions without leaving the app.
Available
See what is sent
Each agent’s folder keeps last-prompt.md: the whole system prompt the app sent on its last turn. Every shipped prompt can be replaced by your own version.
Available
Scripts and other apps
Everything the window can do, a program can do too: the wf command line, or plain HTTP to the running app with the token in runtime.json.
Available
Change how it looks
theme.css on top of the app’s own styles, applied at once. Delete the file to go back. Your agent can do it for you.
06 8 capabilities
Reaches your real tools
An agent that cannot touch your tools is a consultant. These ones can do the work.
Available
An agent’s own browser
Each agent can work in its own off-screen Chrome or Edge profile, with a live view you can watch and take over.
Available
Computer use
On Windows agents prefer real UI automation over raw clicks, and your own mouse and keyboard win. On Linux it runs over AT-SPI, with input on X11 only. On macOS agents get screenshots but cannot control the computer.
Available
Your Android phone
Over ADB, by wireless debugging or USB: read the screen, tap, swipe and type on your own phone.
Available
Connected apps
MCP servers in one file, connectors/mcp.json, in the same format Claude Code and Cursor use. Agents find a server and write the entry themselves; the tools appear within a second.
Available
A page for your apps
Settings › Apps: a catalog of 17 services (Notion, Linear, GitHub, Zapier and more), add by address, add a mailbox, test, switch off, remove. It writes the same files.
Available
Any REST API
No connector needed. An agent calls the API directly with the key as an ID in the header, or writes a small script.
Available
Your own mailbox over IMAP and SMTP with an app password (Gmail, GMX, iCloud and others). A new mail can wake an agent. Outlook and Microsoft 365 are not supported yet.
Available
Telegram
Mirror a chat to Telegram through your own bot and answer your agents from anywhere.
07 4 capabilities
Memory that stays
Your company remembers what it learned, where it found it and when.
Available
Search by meaning
Memory and any folder an agent chooses become searchable by meaning. A small local model does this on your PC, or an embedding provider you pick. Office documents can be read too.
Available
A subconscious beside each agent
A second process with a decision model and a stronger model looks through memory around a turn, notes what people state and tidies long context. What it finds reaches the agent as if the agent had looked it up itself. How reliable this is depends on the decision model.
Available
Forget
An agent forgets on request, and a memory is a file you can delete yourself.
Available
Context ledger
Every part of an agent’s working context has a stable ID. Compaction happens only when needed, and what was hidden or compacted can be restored.
08 4 capabilities
Voice, phone and more PCs
Your company lives on one PC. You can reach it from anywhere you are.
Available
Dictation and calls
Dictate instead of typing. Call an agent inside a chat, optionally with OpenAI Realtime and your own key. Early.
Available
Local speech
Speech-to-text and reading aloud with models that run on your own computer, downloaded when you pick them.
Available
Your phone
Off until you turn it on. Scan a QR code and your PC serves the app to your phone in its browser, over Wi-Fi or Tailscale. No app store, no cloud account.
Available
Several PCs
Pair your computers over the local network, a Remote ID or an address, compare a code, and let agents work on the machine that has the files.
09 7 capabilities
Access, secrets and the brake
Agents may do everything your user account can. Text steers them and helps them; it does not lock anything.
Available
Full access
Agents run without a sandbox and without approval prompts, with your rights. There is no permission gate, no guardian model and no money limit in the app. Read what that means before you install.
Available
The first question
The first thing the AI asks is how far it may go: full access for good, this one search just this once, or less. It is a question, not a technical barrier.
Available
Spending, as a skill
How agents weigh a purchase is written in a skill with reference points and examples: roughly under 20 euros, above that, higher prices. A guide you can edit, not a limit in code.
Available
Secrets by ID
Passwords and keys are encrypted with your operating system’s protection. Agents work with IDs like secret://github, values stay out of prompts, and known values are masked in tool output. Every secret works everywhere; none is tied to a site.
Available
Password import
Import the export file from your browser or password manager once. Your agents then describe and organise the entries by ID, with the values out of their prompts. The export file itself is plain text: delete it afterwards.
Available
Stop all agent work
From the tray, the chat’s working list, the command palette or Ctrl+Alt+Shift+X. Every agent on this PC stops, and stays stopped after a restart until you resume. It does not undo what is done.
Available
Run journal
Tool calls are written down. A call whose effect is unknown after a crash is reported to the agent to check, not repeated blindly.
10 6 capabilities
Yours, on your computer
No account with us, no cloud of ours. Your company is a folder and a database on your own disk.
Available
Local-first
Chats, tasks, memory, agents and the secrets catalog live on your PC. Prompts go to the AI providers you connect.
Available
Backup and restore
A copy of the data folder, without provider logins. Secret values stay in it encrypted for this PC, and agents’ browser profiles with their logged-in sites are copied too. Restoring happens at the next start and keeps a copy of what it replaced.
Available
Your chats as files
A readable copy of every chat sits in the data folder, and Export writes a chat with its messages and files into a folder.
Available
A repair screen
If a start or an update fails repeatedly, the app opens a repair screen instead of leaving you with nothing.
Available
Quiet updates
On Windows and Linux a small icon appears when an update exists; one click installs it and restarts. On macOS the app replaces itself where macOS allows it.
Available
Mac updates without a new download
The Mac app is built to replace itself at an update. A Mac with a version from before the rename gets this one once by hand. It has never run on a real Mac yet, and it falls back to the manual download if macOS refuses.
Ready when you are.
Free download for Windows, macOS and Linux. No account. Bring a ChatGPT plan, a paid Claude plan, an API key or a local model.