FAQ
Is it really free?
Section titled “Is it really free?”Yes. wanna.sh is open source under the MIT licence: no accounts, subscriptions or usage limits. The models are made by others and have their own licences, shown on their Hugging Face pages. Check them before using a model commercially.
Does anything I type leave my computer?
Section titled “Does anything I type leave my computer?”No. Prompts, chats, documents and everything you make stay on your computer. wanna.sh connects to the internet to find and download models and the chat engine, and Code’s web tools fetch the pages and searches the agent asks for.
Does it work offline?
Section titled “Does it work offline?”Yes, once the models you want are downloaded. Finding and downloading new models needs a connection.
Can it answer questions about my own documents?
Section titled “Can it answer questions about my own documents?”Yes. Put them in a Knowledge library (or just attach a long document in Chat), and answers use the passages that match each question, with citations. This is often called RAG. Everything runs on your computer.
How is it different from Ollama, LM Studio or Open WebUI?
Section titled “How is it different from Ollama, LM Studio or Open WebUI?”Those focus on chat. wanna.sh covers chat, pictures, video, voices, transcription, music and coding in one app, rates models for your hardware before you download them, and spreads models across several graphics cards automatically. It also offers OpenAI-compatible endpoints for images, audio and video, not just chat.
Can I use models I already downloaded elsewhere?
Section titled “Can I use models I already downloaded elsewhere?”Not yet: wanna.sh downloads models through Get models so it knows how to run them.
Can I use it with AMD or Intel graphics?
Section titled “Can I use it with AMD or Intel graphics?”Not yet. Without an NVIDIA card, models run on the processor. On a Mac, chat models use Apple’s Metal graphics.
Which chat model should I start with?
Section titled “Which chat model should I start with?”The best one marked Runs great here that has a version of reasonable size. For most computers, a Qwen or Llama model of 3B to 8B parameters is a good balance of speed and quality. For Code, use the biggest model that runs well.
Can several people use it at once?
Section titled “Can several people use it at once?”Yes, through the API or from other devices on your network. See Share on your network.
How do I contribute?
Section titled “How do I contribute?”The code is on GitHub. Issues and pull requests are welcome.