2.3.65 Satellite DeerFlow

DeerFlow

Handle: deerflow
URL: http://localhost:34681

DeerFlow is a community-driven deep research framework that combines LLMs with web search, web crawling, and multi-agent workflows to generate comprehensive research reports.

Key Features:

• Multiple LLM provider support: Works with Ollama, OpenAI-compatible APIs
• Web search integration: Supports Tavily, InfoQuest, DuckDuckGo, and Brave search engines
• Web crawling tools: Integrates with Jina and InfoQuest for content extraction
• RAG support: Connect to RAGFlow, Qdrant, or Milvus for retrieval-augmented generation
• MCP integration: Model Context Protocol server support for enhanced capabilities
• TTS/STT capabilities: Text-to-speech and speech-to-text functionality
• Multi-agent workflows: Complex research tasks distributed across specialized agents
• Python 3.12+: Modern Python backend with FastAPI

Getting Started

# Build images from GitHub repository
harbor pull deerflow

# Start DeerFlow and open in browser
harbor up deerflow --open

• This is a Satellite service (a specialized web UI + backend that integrates with LLM backends like Ollama)
• Default configuration uses DuckDuckGo for search and Jina for web crawling (no API keys required)

Configuration

Environment Variables

Following options can be set via harbor config:

# Web UI port
HARBOR_DEERFLOW_HOST_PORT              34681

# Backend API port (used by the UI)
HARBOR_DEERFLOW_BACKEND_HOST_PORT      34682

# Workspace directory (persistent data)
HARBOR_DEERFLOW_WORKSPACE              ./deerflow

# Frontend image configuration
HARBOR_DEERFLOW_IMAGE                  bytedance/deer-flow-frontend
HARBOR_DEERFLOW_VERSION                latest

# Backend image configuration
HARBOR_DEERFLOW_BACKEND_IMAGE          bytedance/deer-flow-backend
HARBOR_DEERFLOW_BACKEND_VERSION        latest

# Model configuration (used for basic/reasoning/code)
# Note: Must use a model with proper tool calling support (e.g., qwen3:8b, mistral)
HARBOR_DEERFLOW_MODEL                  qwen3:8b

# Search and crawler configuration
HARBOR_DEERFLOW_SEARCH_API             duckduckgo
HARBOR_DEERFLOW_CRAWLER_ENGINE         jina

# Additional settings
HARBOR_DEERFLOW_DEBUG                  false
HARBOR_DEERFLOW_ALLOWED_ORIGINS        http://localhost:34681

Configuration Files

Harbor renders DeerFlow backend configuration at container startup:

• Base fragment: deerflow/configs/deerflow.config.yaml
• Optional overrides (for example, Ollama): additional fragments mounted into /app/configs/*.yaml
• Rendered output inside the container: /app/conf.yaml

If you use the Ollama variant:

harbor up deerflow ollama --open

Harbor will mount an override config fragment that points DeerFlow to Harbor's Ollama endpoint (OpenAI-compatible /v1).

Volumes

DeerFlow persists data in:

• deerflow/data/ - research reports, cache, and generated content
• deerflow/configs/ - Harbor-managed configuration fragments
• deerflow/override.env - optional environment variable overrides

Integration with Harbor

Ollama Integration

Use the Ollama compose variant for automatic local LLM integration:

harbor up deerflow ollama --open

This automatically configures DeerFlow to use your Harbor Ollama instance.

Troubleshooting

Model Requirements

DeerFlow relies heavily on tool calling (function calling) for its multi-agent workflow. If you see "No tool calls in legacy mode - ending workflow gracefully" in the logs, your model doesn't support tool calling properly.

I've tested that at least qwen3:8b and above/newer models work well.

Change the model:

harbor config set deerflow.model "qwen3:8b"
harbor restart deerflow

Check Logs

# All services
harbor logs deerflow

# Backend only
harbor logs deerflow-backend

# Frontend only
harbor logs deerflow

Backend Health Issues

The UI waits for the backend to be healthy before starting. If the UI doesn't start:

harbor logs deerflow-backend
harbor restart deerflow

Configuration Errors

If DeerFlow fails to start, inspect the rendered config inside the backend container:

# While DeerFlow is running, check the rendered config
harbor exec deerflow-backend cat /app/conf.yaml

Links

• Official Repository
• Documentation