Descriptions:
This video walks through a complete from-scratch installation of Agent Zero on a VPS using Docker, with David Ondrej providing the commands and Agent Zero’s founder Jana joining to explain the underlying architecture. The setup uses Docker 29.2.1 and a docker-compose.yaml file (linked as a GitHub Gist) that configures authentication credentials, the OpenRouter API key for model access, and port mappings — with Claude Opus 4.6 as the primary model.
Once Agent Zero is running, Jana demonstrates its knowledge and memory system in detail. The agent stores persistent knowledge files in the /ao/usr/knowledge directory, which is automatically indexed into a vector database for semantic retrieval. A utility model (Kim K2 in the demo) handles memory search and retrieval, while Opus 4.6 generates the final response — a two-stage RAG pipeline built directly into the agent runtime. The video also covers the secret store, which lets agents call external APIs (including image generation via NaraBana Pro through OpenRouter) without ever exposing raw API key values in the context window.
For users who want free model inference, the video previews an alternative approach later in the runtime. The broader setup philosophy — isolated Docker container with a full Kali Linux environment, mapped user folders for persistence, and an agent that can install arbitrary Linux software and expose services on public URLs — positions Agent Zero as a privacy-first, self-hosted alternative to cloud-based AI agents like OpenClaw.
📺 Source: David Ondrej · Published February 07, 2026
🏷️ Format: Tutorial Demo
