Skip to main content

Quick Start

This guide will help you accomplish the following in 5 minutes: log in (optional), search for models, run and chat, manage models, and start the API service.

Prerequisites​

Ensure that the csghub-lite CLI tool and the inference backend are installed. See the Installation Guide.

Step 1: Login (Optional)​

If you need to access private models stored on CSGHub, configure your Access Token in the terminal first:

csghub-lite login

Enter your Access Token from hub.opencsg.com as prompted. You can skip this step for public models.

Step 2: Search for Models​

You can search for models hosted on CSGHub directly from your terminal:

csghub-lite search "Qwen3"

Example output:

Found 34 models (showing top 34):

NAME                                DOWNLOADS   LICENSE      DESCRIPTION
Qwen/Qwen3-0.6B-GGUF               40          apache-2.0   Qwen3-0.6B-GGUF...
Qwen/Qwen3-0.6B                    289229      apache-2.0   ...
...

Tip: Direct local inference requires GGUF format models. It is recommended to include "GGUF" in your search query. If the model is in SafeTensors format, CSGLite will automatically invoke your system Python to convert it to GGUF on the first run.

Step 3: Run and Chat​

Use the run command to automatically download the model (if not already local), start the background inference engine, and enter the interactive chat session:

csghub-lite run Qwen/Qwen3-0.6B-GGUF

The download progress will be shown on the first run. Once complete, the model loads and you can begin chatting:

Loading Qwen/Qwen3-0.6B-GGUF...
Model Qwen/Qwen3-0.6B-GGUF ready. Type '/bye' to exit, '/clear' to reset context.

>>> Hello, introduce yourself
I am an AI assistant focused on helping users solve problems and obtain information.

>>> /bye
Goodbye!

Step 4: Manage Local Models​

# List all downloaded models on your machine
csghub-lite list

# View detailed metadata of a specific model
csghub-lite show Qwen/Qwen3-0.6B-GGUF

# Delete a local model to free up disk space
csghub-lite rm Qwen/Qwen3-0.6B-GGUF

Step 5: Start the API Service Manually​

In addition to being launched automatically during run/chat, you can start the API service independently to serve external applications or custom scripts:

csghub-lite serve

The service listens on localhost:11435 by default. You can test the endpoints using curl:

# 1. Health check
curl http://localhost:11435/api/health

# 2. Chat completion request
curl http://localhost:11435/api/chat -d '{
  "model": "Qwen/Qwen3-0.6B-GGUF",
  "messages": [{"role": "user", "content": "Hello!"}]
}'

# 3. List running models currently loaded in memory
curl http://localhost:11435/api/ps

Next Steps​

  • CLI Reference — Explore advanced parameters and option flags.
  • REST API Reference — Integrate your apps using Ollama and OpenAI-compatible endpoints.
  • Configuration — Change the storage path or connect to a private deployment.