Getting Started with Harbor: A Step-by-Step Tutorial

Getting Started with Harbor

In this step-by-step tutorial, we'll walk through installing Harbor, starting your first LLM stack, and exploring its key features. By the end, you'll have a fully functional local AI environment running in minutes.

Prerequisites

• Docker and Docker Compose installed
• At least 8GB RAM (16GB recommended)
• Linux, macOS, or Windows with WSL2
• Git (for cloning if needed)

Step 1: Install Harbor

The fastest way to install Harbor is via the install script:

curl -fsSL https://raw.githubusercontent.com/av/harbor/main/install.sh | bash

This installs the harbor CLI binary to /usr/local/bin. You can verify the installation with:

harbor --version

Step 2: Start Your First LLM Stack

Starting a basic LLM stack with Open WebUI and Ollama is as simple as:

harbor up

This command:

1. Pulls the latest Ollama and Open WebUI images
2. Generates a Docker Compose configuration with everything pre-wired
3. Starts both services with proper networking
4. Opens Open WebUI in your browser at http://localhost:3000

Step 3: Add Web Search and Voice

One of Harbor's superpowers is adding complementary services to your stack. To add web search and speech capabilities:

harbor up searxng speaches

Harbor automatically connects SearXNG and Speaches to your Open WebUI instance — no configuration needed.

Step 4: Explore Available Services

List all available services with:

harbor list

This shows over 150 services organized by category — LLM backends, frontends, vector databases, MCP servers, agent frameworks, image generation, and more.

Step 5: Use Harbor with Coding Agents

Harbor's launch command lets you use local models from AI coding tools:

# Run Codex against Ollama with Qwen model
harbor launch --backend ollama --model qwen3.5:4b codex

# With web search enabled
harbor launch --web --backend ollama --model qwen3.5:4b codex

Step 6: The Harbor Companion App

Harbor also ships with a desktop companion app built with Tauri. It provides a visual interface for managing services, viewing logs, and monitoring resource usage. The app is available on macOS and Linux.

Advanced: Harbor Boost

Boost adds agentic capabilities to any Harbor backend:

# Start a workflow with Boost modules
harbor launch --workflow quickhop codex

# Run a deep research workflow
harbor launch --workflow deephop --model qwen3.5:4b codex

Troubleshooting

Harbor includes a built-in diagnostic tool:

harbor doctor

This checks Docker availability, port conflicts, and service connectivity. Common issues include port 3000 already in use (Open WebUI) or insufficient Docker resources.

Conclusion

Harbor transforms the complex task of setting up a local AI stack into a single command. Whether you're a developer exploring LLMs, a homelab enthusiast, or a researcher running experiments, Harbor's pre-wired ecosystem of 150+ services saves hours of configuration work. The Apache-2.0 license, active community, and growing service catalog make it one of the most practical tools for self-hosted AI in 2026.