Server · #336 trending MCP repo · ↗ +3 stars this week
Ari-VoiceCommand
DO0OG/Ari-VoiceCommand
Open-source Windows AI voice assistant & desktop automation agent with wake word, STT/TTS, Ollama/local LLMs, MCP, memory, plugins, and agent workflows.
- Stars
- 19
- This week
- +3
- Forks
- 6
- Open issues
- 5
- Latest release
v1.4.0
Documentation
README
🎙️ Ari — Open-Source AI Voice Assistant for Windows
Talk to your PC. Ari listens, answers, and gets things done on your desktop.
Ari is an open-source AI voice assistant for Windows. Call it with a wake word or a hotkey, or click the desktop character, and it listens, answers out loud, and gets things done on your PC. Simple commands run locally right away. Bigger jobs are planned, carried out with tools, and checked step by step.
[!NOTE] Ari runs on Windows 10/11 (64-bit) only.
What Ari can do
| Area | Description |
|---|---|
| Voice conversation | Start talking with a wake word, a hotkey, or a click on the character. Answers are spoken sentence by sentence, and you can interrupt at any time. |
| PC control | Handles commands such as volume, screenshots, and launching apps, as well as multi-step desktop tasks. |
| Fast responses | Common commands are recognized and run on your PC without waiting for an LLM. |
| Model choice | Use local Ollama models, hosted providers, or any OpenAI-compatible server. |
| Speech engines | Speech recognition with Google or offline Whisper. Speech synthesis with Edge, local CosyVoice3, OpenAI, ElevenLabs, Fish Audio, and more. |
| Memory | Remembers what it learns in conversation and brings it up when relevant. Manage it yourself with "remember" and "forget". |
| Extensions | Add features with plugins, SKILL.md skills, and MCP servers. |
| Remote commands | Send commands from Telegram accounts you allow. Off by default. |
| Character | A desktop character shows what Ari is doing and how it feels, and speaks up when something happens. |
| Languages | Korean, English, Japanese |
Things you can say
"What time is it?"
"Set the volume to 30%."
"Take a screenshot."
"Which apps are running?"
"Remember that I prefer short answers."
"Forget what I told you about that preference."
For more complex requests, the agent makes a plan and works through it step by step. What Ari can actually do depends on the features you enable and the model you choose.
Install
Requirements
- Windows 10/11 (64-bit)
- 8 GB RAM recommended
- 4 GB VRAM recommended for running local models on a GPU
- Python 3.11 if you run from source
Installer
Download Ari-Setup-<version>.exe from GitHub Releases and run it. Ari installs to Program Files\Ari by default and keeps settings and history in %AppData%\Ari. To verify the downloaded file, see Release verification and antivirus notes. Signing status: Code signing policy.
Run from source
git clone https://github.com/DO0OG/Ari-VoiceCommand.git
cd Ari-VoiceCommand\VoiceCommand
setup.bat
Ari.vbs
Use setup.bat --with-tts to install local CosyVoice3 as well. If Ari does not start, run Ari.bat to see the error.
Privacy and running locally
With Ollama for the LLM, Whisper for speech recognition, CosyVoice3 for speech synthesis, and local embeddings for memory search, all of that processing stays on your PC. What leaves your PC depends on the providers and features you turn on.
- Remote embeddings and the Telegram integration are off by default.
- Only plugins you approve are loaded. Approved plugins run with the app's permissions and are not sandboxed, so approve only the ones you trust.
Documentation
The documents below are written in Korean.
- User guide — settings, providers and speech engines, skills, automation examples
- Documentation index — all documents, architecture diagram, local decision evaluation results
- Release notes — changes in each version
- For developers: Plugins · MCP server and tools · Autonomous agent · Local decision engine · Themes · Credentials · Contributing
Contributions are always welcome.
Assets & credits
- Default character illustrations by JAraTang: https://www.pixiv.net/users/78194943
- Speech bubble font
Galmuri14by Lee Minseo (quiple), licensed under the SIL Open Font License 1.1: https://github.com/quiple/galmuri
The license text is in VoiceCommand/Galmuri-OFL.txt; include it when you redistribute the font file.
License
Copyright © 2026 DO0OG (MAD_DOGGO).
Ari is released under the MIT License.
Releases