LobeHub Setup Guide: Assemble Your AI Team via the Agent Marketplace (2026)

LobeHub is the largest open-source agent collaboration platform (73K+ GitHub stars), positioning itself as a “Chief Agent Operator” — a general-purpose AI workspace beyond just coding. Its core is the Agent Marketplace + Agent Groups auto-formation + 273K+ Skills / 51K+ MCP Servers ecosystem. This guide walks you through Docker self-hosting in 15 minutes.

Who This Is For

• Already using Claude / GPT, wanting to scale to “agent teams working in parallel”
• Need AI workflows beyond pure coding (writing, research, design, data analysis)
• Prefer self-hosting with your own LLM API keys for long-term cost savings
• Team scenarios: shared agent configs and Skills

TL;DR

# Recommended: interactive setup script
git clone https://github.com/lobehub/lobehub.git
cd lobehub/docker-compose
bash setup.sh    # supports English and Simplified Chinese

# Web UI: http://localhost:3210
# Built-in auth + self-hosted database

Deployment Options

Prerequisites (Docker Compose self-host)

• Docker + Docker Compose
• API key from at least one LLM provider (OpenAI / Anthropic / Gemini / DeepSeek / Ark — any)
• 4GB+ RAM server
• (Optional) S3-compatible storage (AWS S3 / Cloudflare R2 / MinIO), only needed for file upload / knowledge bases

Step 1: Clone and Run the Setup Script

git clone https://github.com/lobehub/lobehub.git
cd lobehub/docker-compose
bash setup.sh

setup.sh is an interactive wizard that asks:

• Cloudflare or direct access?
• Database: built-in Docker Postgres or external?
• Email verification via Resend, or skip for now?
• Enable file upload (requires S3)?

Follow the prompts; the script generates .env and docker-compose.yml automatically.

💡 Friendly for Docker beginners: the script supports interactive Q&A in English / Chinese, much easier than hand-editing .env.

Step 2: Start the Services

docker compose up -d

# Watch logs to confirm all services started
docker compose logs -f

Visit http://localhost:3210.

Step 3: Configure LLM API Keys

In the web UI:

1. Settings → AI Providers
2. Pick a provider (OpenAI / Anthropic / Gemini / custom OpenAI-compatible endpoint)
3. Fill API Key + Base URL (for domestic gateways like Ark, use https://ark.cn-beijing.volces.com/api/coding)
4. Test connection → Save

⚠️ Never commit API keys to git — .env is gitignored, but double-check if you copy keys to other config files.

Step 4: Pick Agents from the Marketplace

LobeHub’s edge is its ecosystem. Visit Discover → Agents:

• 273K+ Skills (as of 2026-05)
• 51K+ MCP Servers
• Agents cover coding, writing, data analysis, design, more

Pick a solid one (e.g., “Frontend Developer”), click Add to My Workspace. It’ll appear in your agent selector next time you start a new chat.

Step 5: Use Agent Groups for Auto Team Formation

For complex tasks (e.g., “build a login page”) that a single agent can’t handle, use Agent Group:

1. My Workspace → New Group
2. Describe the goal (“build a login page with form validation”)
3. LobeHub’s CAO (Chief Agent Operator) automatically selects a fitting combination from your Workspace (Frontend Agent + Test Agent + Code Review Agent)
4. Agents collaborate within the Group; CAO aggregates the final output

This is LobeHub’s core differentiator from single-agent tools.

Persistent Memory and Continuous Learning

Each agent has its own memory store that learns your preferences. For example:

• First time you ask the Frontend Agent to write a component, it asks “What CSS approach?” — you say Tailwind
• Next time it defaults to Tailwind without asking
• Synced across sessions and devices (self-host uses your Postgres)

Common Pitfalls

1. First setup.sh run is slow → images pulled from GHCR; in China, configure an Alibaba mirror or use a proxy for docker login ghcr.io
2. API key configured but agent doesn’t reply → check docker logs; common cause is wrong Base URL (OpenAI-compatible endpoints must end with /v1 or omit it)
3. Multi-agent Group blows up API costs → assign cheaper models (e.g., mini) to peripheral agents, flagship only for final review
4. Cloud Pro $20 vs self-host → Pro’s compute credits with optimized routing have higher ROI for heavy users; light users save more with self-host + BYO API

Comparison with Other Platforms

See the 2026 Agent Collaboration Platform Guide for full comparison.

Related

• LobeHub product page
• 2026 Agent Collaboration Platform Roundup
• Multica setup guide (open-source coding collaboration)
• MCP intro (understanding LobeHub’s 51K+ MCP servers)

Verified through 2026-05-19. Ecosystem numbers from LobeHub Marketplace and lobehub.com/mcp.