Dictation
Speak your tasks instead of typing — hold Ctrl+Space, speak, release. Recognition runs locally on your computer.
With dictation you simply speak your tasks. The text lands where you are:
- in a CLI's terminal (Claude Code, Codex, Kimi, Gemini, Hermes) — as if typed. You press Enter yourself, so you can check the text first.
- in a text field — orchestrator chat, threads, task bar — at the cursor.
How it works
- Click into the terminal or field the text should go to.
- Hold Ctrl+Space and speak. The golden orb appears at the bottom — it pulses with your voice, with time and target below.
- Release — the text is recognised and inserted.
Esc cancels. You can also use the microphone button next to the input fields: hold it, or click once and click again to insert.
First-time setup
On your first dictation OpenCognit asks whether to download the speech model (about 670 MB, once). After that, recognition runs fully offline — no account, no upload, no cost.
Languages
The model (NVIDIA Parakeet TDT 0.6B v3, CC-BY-4.0) understands English, German and 23 more European languages and detects the language by itself.
If nothing happens
- Ctrl+Space does nothing: some Linux input methods (ibus, fcitx) use these keys to switch languages. Use the microphone button instead.
- "No access to the microphone": check that a microphone is connected and not in use by another program.
- "Too short": keep the keys pressed while you speak.