Player not loading? Watch on YouTube
This tutorial follows AnythingLLM setup on a Windows x64 desktop, then a Docker deployment on a Hostinger KVM2 VPS. The speaker also identifies macOS and Linux downloads. Settings cover model providers, the bundled LanceDB vector database, embeddings and text chunking.
The demonstration uses OpenRouter with an API key. The speaker suggests Ollama or LM Studio to run models locally, so the demonstrated configuration should not be treated as offline. The sources disagree on the DeepSeek version: the description names V3 Flash, while the transcript says V4 Flash.
For RAG, the speaker uploads a 41-page Tesla PDF and asks about automotive leasing revenue. Agent settings include document creation, chart generation and web search. File access requires permission for a selected folder; the demonstration checks which files the agent can access. Custom skills, an agent flow and JSON-based MCP configuration also receive a walkthrough.
The self-hosted setup uses Hostinger's Docker catalog. The speaker configures the model provider again, creates a Telegram bot through BotFather, enters its token and approves a pairing request before sending a message from a phone. The VPS provides an instance that stays available away from the laptop. Performance and cost comparisons are the speaker's observations, rather than general benchmarks.
