Perplexica (Vane): Docker setup and Perplexity comparison

The tutorial introduces Vane, formerly Perplexica, as an open source research tool with a workflow similar to Perplexity: search the web, collect sources, then generate an answer with a language model. The presenter describes SearXNG as its search backend and explains that users can connect either a local LLM or a cloud provider.

Setup starts with Docker Desktop, with Mac and Windows download options discussed. The presenter runs a command from the repository to pull and start the container, opens the browser interface, and adds a model connection. Although Ollama appears as a connection option, the demonstration uses cloud APIs rather than showing how to run models locally.

The first attempt uses a key created in Google AI Studio and selects Gemini 3 Flash. The presenter also discusses choosing an embedding model, mentioning MiniLM. After an apparent Gemini API issue, the presenter replaces the connection with OpenAI and receives a response following a web search. The cause of the Gemini failure is not established.

For machines with 8GB of memory, the presenter recommends smaller models and warns that larger models may respond slowly. This is a self-hosted interface walkthrough, not an offline demonstration: the research workflow searches the internet, and the successful example uses a cloud model.