8. Guides

Harbor Guides

Practical guides for building and operating local AI setups with Harbor. These pages focus on common workflows that combine multiple Harbor services rather than documenting a single service in isolation.

• Local LLM Stack with Docker Compose
  Build a local LLM stack with Harbor's Ollama and Open WebUI services, Docker Compose orchestration, and optional cross-service additions such as SearXNG, Speaches, and ComfyUI.
• Ollama + Open WebUI + SearXNG Local Web RAG Setup
  Add self-hosted web search to an Ollama and Open WebUI stack with Harbor's SearXNG integration.
• Run Coding Agents with Local LLMs
  Use Harbor Launch to run Codex, Claude Code, OpenCode, and other host coding tools against local OpenAI-compatible backends.
• Harbor vs Manual Docker Compose for Local AI
  Compare a manual Docker Compose local AI stack with Harbor's service selection, cross-service integrations, configuration, CLI/App workflows, Launch adapters, and harbor eject.
• OpenAI-Compatible Local LLM Backends
  Choose and swap Harbor backends such as Ollama, llama.cpp, vLLM, Docker Model Runner, MLX, and oMLX for local OpenAI-compatible frontends, clients, and Harbor Launch workflows.
• Self-Hosted AI Stack for Developers
  Plan a private AI stack with Harbor across chat, coding, RAG/search, tools, image generation, voice, workflows, benchmarks, and observability.
• Run Hermes Agent with Harbor
  Run Hermes Agent with Harbor local LLM backends through harbor launch or the containerized Hermes service/API server.
• Run MLX on Apple Silicon with Harbor: Local LLMs and HuggingFace Models
  Install and run Apple's MLX framework on M-series Macs, discover and pull optimized models from HuggingFace (mlx-community), and integrate MLX into full Harbor stacks with Open WebUI, SearXNG, and coding agents.
• Docker Model Runner Local LLMs with Harbor on Mac, Linux, and Windows
  Use Docker's official Model Runner as a backend on every supported platform. Find GGUF models via docker model search --source huggingface, pull with Harbor's unified interface, and run complete local LLM workflows.