You think.
They build.
Run Claude Code, Codex, Gemini CLI, Kimi and Hermes side by side — locally, with the subscriptions you already have.
- Free to start
- Pro one-time, no subscription
- No account, no telemetry
How it works
I · Install
Download for Linux or Windows. The setup assistant finds your installed CLIs and helps you sign in.
II · Open projects
Click plus, pick a folder, pick a CLI — the agent starts right in that project. As many projects as you like, side by side.
III · Let them work
Terminals in the grid or threads with approvals: you see every tool step, risky commands wait for your click.
IV · Keep track
The Cosmos shows live which agent reads or changes which file — and warns when two touch the same one.
V · Review & commit
Read the diff, comment lines, send them back to the agent — then commit and push, all inside the app.
Your CLIs. Your subscriptions.
OpenCognit drives the official coding CLIs you already use and pay for. No API keys, no extra model, no per-call fees. Installed CLIs are detected automatically — and each one updates itself as usual.
Claude Codeterminal · thread · conductor
Kimi Codeterminal · thread · conductor
Hermesthread · conductor · 60+ models
Codexterminal · review
Gemini CLIterminal · review
gitcommit · push · branchesLook inside the app.
Real screenshots of OpenCognit — six agents, five CLIs, one window.
Nothing risky slips through
Before an agent pushes, deploys or deletes, OpenCognit stops and asks — even for commands the agent runs on its own. Secrets in commits are blocked outright.
Screenshots with demo data.
Just say what you want.
Hold Ctrl+Space, speak, let go — your words land in the terminal or field you are in. Recognised on your own computer.
Hold, speak, release
Works in every agent terminal — Claude Code, Codex, Kimi, Gemini, Hermes — and in the orchestrator chat. You press Enter yourself.
Offline, no account
Speech recognition runs locally (NVIDIA Parakeet). No upload, no subscription, no credits. One-time model download, then it works offline.
German, English and 23 more
The language is detected automatically, even when you switch mid-sentence.
Included — no extra subscription.
✻ Welcome to Claude Code
❯
Animation — no microphone is used on this page.
More agents.
More control.
Starting several agents at once is something almost every tool can do today. OpenCognit makes sure they only do what they should — and that you always see what they are doing and what it costs.
Guardrails out of the box
Commands from Claude Code, Codex and Gemini run through your rules first: block, ask or warn. Before every commit, OpenCognit scans for keys and passwords.
Your CLIs, as they are
Every agent in its real terminal, with its own commands, plugins and logins — or as a thread where questions and approvals arrive as cards. All projects in one grid.
Orchestrator that reports back
Say what you want: the orchestrator puts the team together, hands out the tasks and comes back on its own as soon as everyone is done.
Usage in euros
Tokens of every CLI per project and model, with API equivalent and subscription savings — read from local logs, nothing leaves your machine.
Schedule
Agents work at set times, for example running the tests every morning and writing a short report. Anything risky waits for your approval.
Git & review loop
Review changes, comment lines and send them back — commit, discard, switch branches, pull and push without leaving the app.
Editor, terminal & search
Edit files, open a shell (Ctrl+J), search the whole project (Ctrl+Shift+F) — right next to your agents.
Several accounts per CLI
Work and private logins side by side. When a quota runs out, you see it — and switch to the next account.
Cosmos
Your project as a star system. When an agent reads or changes a file, a beam of light shows it — and conflicts become visible early.
In control
instead of flying blind.
Without OpenCognit
With OpenCognit
Closed source.
Open about everything else.
OpenCognit runs agents with access to your code and your terminal. Here is exactly what it does – and how you can check it yourself.
Signed updates
Every update is signed with our release key. The app refuses anything that is not.
Nothing phones home
No telemetry, no account. The handbook lists every request the app makes – check it with your firewall.
Your secrets stay put
Agents need your OK before they read credentials or upload files. Their terminals get no secrets from your environment.
Licence works offline
Your key is checked on your computer. No server of ours has to be online – ever.
Pay once.
Keep it.
No subscription next to your AI subscriptions. No account, no telemetry. Try it for free, buy once — the version you bought keeps working forever.
- Up to 4 agents at the same time
- 2 projects (workspaces)
- All features incl. Cosmos, editor, threads and dictation
- No account, no time limit
- Unlimited agents and projects
- 12 months of updates included
- Linux and Windows
- License key right after payment
- Checked offline — no account needed
- Another 12 months of updates
- Your version keeps working without it — forever
- The app tells you when a new version is available
AI costs are not included — you use your own CLI subscriptions (Claude Code, Kimi, Codex, Gemini, Hermes). OpenCognit is the control center, not a model.