Player not loading? Watch on YouTube
This tutorial covers a self-hosted MaxKB setup on Ubuntu, followed by model configuration and a document-based assistant. The speaker describes MaxKB as an open source platform with retrieval augmented generation and AI agent workflows. Docker is required for the installation shown; the virtual environment is optional. After the container starts, the speaker opens localhost port 8080, signs in and resets the password.
The demonstration introduces Ollama as a local LLM connection and explains the model types available in MaxKB, including text generation, vision, embeddings and reranking. The speaker initially says the example will use API-based models, so the walkthrough should not be read as proof of a fully offline setup. The machine has an RTX 6000 with 48 GB of VRAM, but the video does not establish that hardware as a minimum requirement.
A PDF becomes the source for a knowledge base. The speaker creates a simple application, selects a model and edits its system prompt. An initial question produces an unrelated answer; after attaching the data source, the speaker reports that the answer comes from the uploaded material. The interface also includes MCP tools and publishing controls.
The review notes interface errors and generated text that falls back to Chinese. The speaker recommends Ollama for testing and vLLM for production, and argues that demanding agent workflows still need large hosted models. These are the speaker's assessments, rather than general performance guarantees.
