Ollama와 함께하는 지역 모델
Download and run open models on your own computer, free and private, with the built-in Local Models manager.
9xChat runs open models such as Llama, Qwen and Gemma on your own computer through Ollama. A local model costs nothing per message, needs no API key and no credits, and keeps every prompt on your machine.
Local models run on desktop (Windows, macOS and Linux) on every plan. Mobile uses cloud models.
Set up in three steps
-
Install Ollama from ollama.com. If it is missing, 9xChat says Ollama is not installed and offers Install Ollama; if it is installed but stopped, it says Ollama is not running.
-
Open Library → Models and find Locally Hosted.
-
Choose Browse & Download to get a new model, or Scan to add the models you already pulled.
Choosing a model that fits
Browse & Download opens the Local Models manager. It suggests models for chat, coding and knowledge search, such as Gemma 2 2B, Llama 3.2 3B, Llama 3.1 8B, Qwen2.5 Coder 7B, Nomic Embed Text and MixedBread Embed Large, and rates each one against your computer:
| Badge | 의미 |
|---|---|
| Should run well | Fits comfortably in your memory |
| Tight / may be slow | Runs, but answers may be slow |
| Likely won't fit | Too large for this computer |
A Low disk space warning appears before a download that would fill your drive.
Any other model
Pull any model takes any Ollama model name, such as qwen2.5:14b, or a GGUF model from Hugging Face in the form hf.co/<repo>-GGUF:Q4_K_M.
A remote or custom Ollama
Ollama host defaults to http://localhost:11434. Point it at a custom port, or at Ollama on another machine on your network, and leave it blank to go back to the default.
Using a local model
Local models appear in the chat model picker under Local Models, next to your cloud models, so you can switch per message: a local model for anything private, a frontier model when you need the strongest answer.
-
Knowledge search: set Settings → Default models → Search your files by meaning to Local (Ollama) to index your files on your own computer, for free.
-
Tools, agents and Code: for agentic work such as web search, tools and the Code workspace, a cloud model built for tool use gives the most reliable results.
개인 정보 보호
- With
localhost, every request stays on your computer. - With a remote host, requests travel over your network to that machine.
- Web search, read aloud and dictation use their own online services even when your chat model is local. See Privacy and data flow.
If it does not connect
- Confirm Ollama is running, then choose Scan again.
- For a remote machine or Docker, check the Ollama host address and that Ollama accepts connections from other machines.
- If answers are slow, pick a model marked Should run well.