My_AI

What does My_AI do?

My_AI describes itself as "Your private AI assistant, running 100% on your computer." It is a Python app with a pywebview window and an Ollama back end, so the model runs on the device. The README lists nine agents (research, code, writing, data, planning and so on) that can be chained through a visual workflow builder, and the output can be saved as a Word document, a PDF, an Excel file or a PowerPoint deck.

The UI is translated into 12 languages. A codebase RAG mode reads a repository and answers questions about the code. Web search and a mobile sync are optional and go over the network; the maintainer's README says the rest works offline once the model is downloaded. A setup wizard reads the machine's RAM and GPU and picks a model size from Qwen 3.5 2B, 4B or 9B.

Key facts

• Licence: MIT.
• Platforms: Windows, macOS, Linux. Python 3.10 or later is required.
• Model runtime: Ollama runs the model on the device.
• Models: Qwen 3.5 2B (4 GB RAM floor), Qwen 3.5 4B (8 GB), Qwen 3.5 9B (16 GB). The setup wizard picks one based on the detected hardware.
• UI languages: 12, listed in the README.
• Agents: 9 built-in agents, chainable through a visual workflow.
• File output: Word, PDF, Excel, PowerPoint, generated from the chat.
• Code mode: a codebase RAG that reads a repository and answers questions about its code.
• Offline by default: once the model is downloaded, the README says the local features work without internet.
• Online features: web search and the mobile companion need internet.
• UI: a native window through pywebview, so no browser is needed.
• Stars when read: 13 (small project).

What does it replace, and where does it fall short?

My_AI replaces ChatGPT for a reader who wants a private assistant that runs on a laptop, generates office files from the chat and keeps the model on the device.

Where it falls short: Python 3.10 and Ollama have to be installed first, so the install is more work than a disk image. The bundled model family is Qwen 3.5; other open-weights families (Llama, Mistral, DeepSeek) are not shipped, although Ollama itself can run them if a user sets them up. Web search and the mobile companion go over the network, so the "100% on your computer" line applies only to the default local path. The project is small (13 stars when read), so the community support base is narrow.

How does My_AI run?

Install Python 3.10 or later and Ollama. Clone the repository and install the Python dependencies with pip install -r requirements.txt. Run python main.py; the setup wizard reads the machine's RAM and GPU, then downloads the right Qwen 3.5 size (2B, 4B or 9B).

Once the model is ready, the pywebview window opens the assistant. Pick an agent from the nine built in, or chain agents through the workflow builder. Save a reply as a Word document, PDF, Excel file or PowerPoint deck from the Export menu. The codebase RAG reads a selected folder and answers questions about the code.

Who is My_AI for?

A Windows, macOS or Linux user who already has Python and is comfortable with a pip install, who wants a private assistant with office file export and visual workflows. A reader who wants a one-click installer should look at Jan or LobeHub. A reader who wants a chat UI with many model families should pair Open WebUI with Ollama.

What limits does the README state?

• Python 3.10 or later and Ollama are required on the system before install.
• The bundled model family is Qwen 3.5; other model families need manual setup in Ollama.
• Web search and the mobile companion need internet and are optional.
• The setup wizard picks a model by detected RAM and GPU; a user can override the pick.
• The README gives no measured latency or accuracy figure for any model size.

Questions people ask

Is My_AI open source?

Yes. The LICENSE file at the repository root is MIT, an OSI-approved licence. The code is Python with pywebview. Ollama is MIT. The Qwen 3.5 models have their own licences on Hugging Face: Alibaba's Qwen team publishes the model cards and licence terms there.

Does My_AI work offline?

Yes, in the default configuration. Once Ollama has downloaded the Qwen 3.5 model, the chat, agents and workflows run on the device. The README's own line is "Once the models are downloaded, everything except web search and mobile access works offline." Web search and the mobile companion call external services and need internet.

Can My_AI be used commercially?

Yes for the application code, which is MIT. Ollama is MIT. The Qwen 3.5 model licences on Hugging Face have their own terms, and most Qwen open-weights models allow commercial use below a user count, with a separate licence above that. A commercial user should read the model card for the exact size before shipping.

How does My_AI compare with ChatGPT?

ChatGPT runs on OpenAI's hosted models and keeps the conversation on OpenAI's servers. My_AI runs Qwen 3.5 on the user's laptop through Ollama, keeps files and memory on the laptop, and ships in a native pywebview window with nine agents and office export. The trade is: ChatGPT's hosted quality against My_AI's on-device privacy and office integration. The open-source alternatives to ChatGPT page sets it beside LobeHub, Jan and LibreChat.

What are the nine agents?

The README lists research, code, writing, data, planning, and four others. Each is a prompt plus a set of tools, and they can be chained through the visual workflow. The result is turned into a chat reply, or saved as a Word, PDF, Excel or PowerPoint file. The list is set in the repository rather than downloaded from a store, so the agents are the ones the maintainer ships.

Sources

• My_AI README and LICENSE: github.com/gonicolas12/My_AI, read 2026-11-09.