How to Self-Host DeepClaude: A Step-by-Step Setup Guide

Want to run your own dual-model AI inference server combining DeepSeek R1's reasoning with Claude's creative output? This step-by-step guide will walk you through deploying DeepClaude on your own hardware (or VPS) using Docker.

Prerequisites

Step 1: Pull the Docker Image

docker pull erlichliu/deepclaude:latest

The image is relatively lightweight and includes the Rust binary, web frontend, and all dependencies.

Step 2: Configure Your API Keys

DeepClaude uses a config.toml file for all configuration. Create it in a working directory:

[api_keys]
deepseek = "sk-your-deepseek-key-here"
anthropic = "sk-ant-your-anthropic-key-here"

[models]
reasoning = "deepseek-reasoner"
creative = "claude-sonnet-4-20250514"

[server]
host = "0.0.0.0"
port = 8080

Step 3: Run with Docker

docker run -d \
  --name deepclaude \
  -p 8080:8080 \
  -v $(pwd)/config.toml:/usr/local/bin/config.toml \
  erlichliu/deepclaude:latest

Step 4: Verify It's Running

curl http://localhost:8080/v1/health

You should see a JSON response confirming the server is ready. Now open http://localhost:8080 in your browser to access the chat interface.

Step 5: Using Docker Compose (Recommended)

For a more production-ready setup, create a docker-compose.yml:

version: '3.8'
services:
  deepclaude:
    image: erlichliu/deepclaude:latest
    restart: unless-stopped
    ports:
      - "8080:8080"
    volumes:
      - ./config.toml:/usr/local/bin/config.toml
      - ./data:/data

Then run:

docker compose up -d

API Usage Examples

DeepClaude supports both OpenAI-compatible and Anthropic-compatible API formats. Here's a basic curl example:

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepclaude",
    "messages": [
      {"role": "user", "content": "Write a Python function to merge two sorted lists"}
    ],
    "stream": true
  }'

For streaming, you'll receive SSE events: first the R1 reasoning trace (as reasoning events), then Claude's response (as standard content events).

Troubleshooting Tips

• Connection refused: Make sure the container is running and ports are properly mapped. Check with docker ps.
• API key errors: Verify your keys are correct and have sufficient credits. The server logs will show 401 errors if keys are invalid.
• Out of memory: LLM inference is memory-intensive. Ensure your system meets the recommended 8 GB RAM.
• Slow responses: Both APIs have latency. The R1 reasoning step typically takes 3-10 seconds depending on complexity.