Player not loading? Watch on YouTube
Uploaded on January 8, 2025, this Chinese-language tutorial demonstrates a Windows setup for a self-hosted FastGPT assistant using Ollama and Mistral. The supplied material gives no recording date or software release numbers. Its version context is WSL2 and a Mistral model selected by name, so the deployment steps describe the setup shown at that time.
The speaker explains a Docker stack containing FastGPT, MongoDB and PostgreSQL. OneAPI connects the application to model providers, including Ollama. The architecture discussion distinguishes language models from embedding models used for knowledge bases, though the walkthrough focuses on a basic chat application.
Setup begins with WSL2 and Ubuntu, followed by Ollama for Windows and a Mistral download. The presenter installs Docker Desktop, enables WSL integration and downloads FastGPT's configuration and pgvector Compose files. After setting the frontend address to the machine's IP on port 3000, the presenter starts the containers and adds an Ollama channel in OneAPI.
A useful troubleshooting example follows: configuring OneAPI alone does not make Mistral appear in FastGPT's model selector. The presenter also edits the LLM model entries in config.json and restarts the services. The tutorial ends with a chat test, publication of the assistant and a link that lets other users chat without logging in. It also shows webpage embedding options.
