Qwen3 1.7B
For everyday questions, drafting messages, explaining ideas and finding a starting point.
- Model family
- Qwen, by the Qwen team at Alibaba
- Neo’s version
- Q8_0 · a compact GGUF model file
About Local AI
Get to know
your AI.
Neo brings a few useful AI models into one simple app on your computer. Here’s what’s inside, how it works, and what you can do with it.
01 / The models
Three assistants.
Different strengths.
Neo is the app that brings them together. Each assistant uses an existing model developed by its credited publisher.
Qwen2.5-Coder 1.5B
For understanding code, working through a small programming task and drafting simple functions.
- Model family
- Qwen, by the Qwen team at Alibaba
- Neo’s version
- Instruct · Q4_K_M GGUF
SmolVLM 500M Instruct
For describing a picture and answering simple questions about what is visible. This model reads images; it does not generate them.
- Model family
- SmolVLM, by Hugging Face
- Neo’s version
- Instruct · Q8_0 GGUF + vision projector
“B” means billion parameters; “M” means million. These describe model size, not a guarantee of accuracy. The compact versions help Neo fit on everyday computers.
02 / How it was created
A local app,
built around open tools.
Neo Local AI is a project by Omar Imran. Its interface and local services connect existing AI models into one experience. Neo does not train or fine-tune the underlying models.
- 01
An interface you can open.
The interface is built with HTML, CSS and JavaScript. On Mac, a native Swift window makes it open like a desktop app. The Windows preview opens the local interface in a browser.
- 02
An engine on your computer.
Python handles the app’s local requests, while llama.cpp loads and runs the chat, code and vision models. Neo switches between models to keep memory use manageable.
- 03
Models that work offline.
Setup downloads model files from Hugging Face. After that, the installed models can answer locally without an AI subscription or an internet connection. Neo does not browse the web for current information.
03 / The features
What you can do.
Start with a conversation. Choose the assistant that fits your task, then use the other tools when you need them.
Chat, code & pictures
Ask questions, get writing or coding help, and attach a picture for the vision model to explain.
Appearance
Switch between dark and light. In the 2.0 workspace, open Settings at the top right for theme and accent choices.
Private memory
Connect your own Obsidian vault to keep useful notes locally. If that vault uses a sync service, its existing sync settings still apply.
THE 2.0 DEVELOPMENT WORKSPACE
More tools, when you need them.
The features below are part of Neo’s 2.0 workspace. The public Mac download is currently 1.3, and the Windows download is a 1.3 preview. These newer tools are not included in those downloads yet.
Model hub
See your installed models and choose optional downloads: Qwen3 0.6B for a smaller model, or Qwen3 4B for a larger one that needs more memory.
Creative studio
Generate pictures using Stable Diffusion 1.5 through a local ComfyUI engine. Analyze a video through up to four sampled frames; this does not include audio. Video generation needs a separate compatible workflow and models.
Plugins
Choose reusable instructions for writing, studying or planning, or create your own. These prompt plugins do not run code or automatically access your files.
Integrations
Connect supported local tools such as Ollama and ComfyUI, and manage your Obsidian memory connection.
Local API
Open Settings → Local API to connect your own software on the same computer. It is off by default, requires a separate key, and does not include your private memory.
Saved conversations
Return to previous chats through History, or export a conversation as a Markdown file to keep a copy.