Kostenloser lokaler KI-Assistent: So kombinieren sich OpenClaw und DeepSeek

DeepSeek is a powerful open-source AI model family (1.5B to 671B parameters) with exceptional reasoning capabilities. OpenClaw is a viral open-source AI assistant (170K+ GitHub stars) that runs entirely locally. Combine the two via Ollama and you get a free, privacy-focused AI assistant that rivals paid alternatives — no API costs, no subscriptions, complete control.

Building a personal AI assistant has never been more accessible. Between API costs, subscription plans, and privacy concerns, developers need a clear path to getting started with local AI capabilities.

This guide walks you through setting up DeepSeek — the open-source model family from DeepSeek AI — with OpenClaw, a viral open-source AI assistant that runs entirely on your hardware.

The best part? Both DeepSeek and OpenClaw are free to use. No credit card. No subscription. No data leaving your machine.

By the end of this guide, you will have:

Whether you're a developer looking to automate tasks, a hobbyist exploring local AI, or a business seeking privacy-first AI solutions, this setup delivers enterprise-grade capabilities at zero cost.

DeepSeek has emerged as one of the most capable open-source AI model families in 2026. Here's what makes it stand out:

Exceptional reasoning — DeepSeek-R1 achieves performance approaching leading models like OpenAI O3 and Gemini 2.5 Pro on reasoning tasks. It's particularly strong in mathematics, coding, and complex problem-solving.

Model variety — DeepSeek offers models for every use case:

Hybrid reasoning — Like Qwen3, DeepSeek-V3.1 supports both thinking mode (Chain-of-Thought reasoning) and non-thinking mode (direct answers), letting you choose based on your task.

Cost efficiency — DeepSeek models are open-source and free to run locally. You only pay for the hardware.

OpenClaw (formerly Clawdbot/Moltbot) is an open-source AI agent with 170,000+ GitHub stars.

DeepSeek's powerful reasoning plus OpenClaw's agentic capabilities creates a free, private AI assistant that rivals paid alternatives:

Pro tip: Start with the 7B model if you have 16GB RAM. You can always scale up later.

Ollama is the bridge that lets you run DeepSeek locally. It handles model downloading, memory management, and inference serving.

# Using Homebrew (recommended) brew install ollama # Or using the install script curl -fsSL https://ollama.ai/install.sh | sh Enter fullscreen mode Exit fullscreen mode Linux # Using the install script curl -fsSL https://ollama.ai/install.sh | sh # Or download the binary directly sudo curl -L https://ollama.ai/download/ollama-linux-amd64 -o /usr/bin/ollama sudo chmod +x /usr/bin/ollama Enter fullscreen mode Exit fullscreen mode Windows Download and run the installer from ollama.ai.

ollama --version Enter fullscreen mode Exit fullscreen mode You should see output like ollama version 0.5.0 or similar.

Ollama runs as a background service. It should start automatically, but you can verify:

# Check if Ollama is running ollama list # If not running, start it ollama serve Enter fullscreen mode Exit fullscreen mode Step 2: Pull and Run a DeepSeek Model Pull DeepSeek-R1 (Recommended) DeepSeek-R1 is the flagship reasoning model. For most users, start with the 7B or 8B model:

# Pull the 7B model (recommended for most users) ollama pull deepseek-r1:7b # Or pull the 8B model for slightly better performance ollama pull deepseek-r1:8b # For more powerful hardware, try the 14B model ollama pull deepseek-r1:14b Enter fullscreen mode Exit fullscreen mode Pull DeepSeek-V3 (General Purpose) If you need a general-purpose model rather than a reasoning-focused one:

# Pull DeepSeek-V3 (requires significant RAM) ollama pull deepseek-v3:671b Enter fullscreen mode Exit fullscreen mode Pull Distilled Models (Low Resource) For systems with limited RAM, distilled models offer good reasoning at smaller sizes:

# Pull distilled models based on Qwen architecture ollama pull deepseek-r1:1.5b ollama pull deepseek-r1:14b Enter fullscreen mode Exit fullscreen mode Run the Model # Interactive chat mode ollama run deepseek-r1:7b Enter fullscreen mode Exit fullscreen mode Type your message and press Enter. Type /exit to quit.

Here's how to use DeepSeek programmatically:

import requests url = " http://localhost:11434/api/generate " payload = { " model ": " deepseek-r1:7b ", " prompt ": " Explain what DeepSeek R1 is in one sentence ", " stream ": False, } response = requests. post ( url, json = payload ). json () print ( response [ " response " ]) Enter fullscreen mode Exit fullscreen mode Test Your Ollama API with Apidog Before integrating with OpenClaw, test your DeepSeek setup using Apidog. This is especially useful for debugging and verifying your API endpoints work correctly.

{ "model": "deepseek-r1:7b", "prompt": "Hello, world!", "stream": false } Enter fullscreen mode Exit fullscreen mode Apidog's visual interface makes it easy to test your Ollama API responses and debug any issues before connecting to OpenClaw. You can also save this request to test different prompts and configurations.

from ollama import Client client = Client () output = client. chat ( model = " deepseek-r1:7b ", messages = [{ " role ": " user ", " content ": " Write a hello world in Python " }] ) print ( output [ " message " ][ " content " ]) Enter fullscreen mode Exit fullscreen mode Step 3: Install OpenClaw Quick Installation # Using npx (no installation needed) npx openclaw # Or using the installation script curl -fsSL https://openclaw.ai/install.sh | bash Enter fullscreen mode Exit fullscreen mode Initial Setup Run OpenClaw for the first time:

npx openclaw Enter fullscreen mode Exit fullscreen mode

This will guide you through initial configuration:

# Check OpenClaw status openclaw status Enter fullscreen mode Exit fullscreen mode Step 4: Connect DeepSeek to OpenClaw Now the magic happens — we connect DeepSeek as the brain of your OpenClaw assistant.

OpenClaw natively supports Ollama. Configure it to use DeepSeek:

# Set OpenClaw to use Ollama with DeepSeek-R1 ollama launch openclaw --model deepseek-r1 # Or specify a different model size ollama launch openclaw --model deepseek-v3.1 Enter fullscreen mode Exit fullscreen mode Method 2: Environment Configuration Set environment variables for more control:

# Configure Ollama endpoint export OLLAMA_HOST = http://localhost:11434 # Set the model export OLLAMA_MODEL = deepseek-r1 Enter fullscreen mode Exit fullscreen mode Method 3: Configuration File Create or edit ~/.openclaw/config.yaml:

models: default: ollama/deepseek-r1:7b ollama: host: http://localhost:11434 model: deepseek-r1:7b temperature: 0.7 top_p: 0.9 Enter fullscreen mode Exit fullscreen mode Test the Integration # Test that OpenClaw is using DeepSeek openclaw models status Enter fullscreen mode Exit fullscreen mode You should see output confirming DeepSeek-R1 is active.

Now you can chat with DeepSeek through any connected platform:

The response will come from DeepSeek running locally!

Fine-tune your DeepSeek + OpenClaw setup with these options.

# In config.yaml ollama: temperature: 0.7 # 0.0 = focused, 1.0 = creative top_p: 0.9 # Nucleus sampling top_k: 40 # Token selection Enter fullscreen mode Exit fullscreen mode Context Length Adjust for longer conversations:

ollama: context_size: 4096 # Increase for longer context Enter fullscreen mode Exit fullscreen mode System Prompt Customize DeepSeek's behavior:

ollama: system_prompt: | You are a helpful coding assistant. You provide clear, concise code examples. You explain concepts in simple terms. Enter fullscreen mode Exit fullscreen mode Switch Between Models You can easily switch between different DeepSeek models based on your needs:

# Switch to the 14B model for more capability openclaw models set ollama/deepseek-r1:14b # Switch to V3 for general tasks openclaw models set ollama/deepseek-v3:671b # Switch back to 7B for speed openclaw models set ollama/deepseek-r1:7b Enter fullscreen mode Exit fullscreen mode Testing Your AI Assistant Test via Ollama Directly # Test DeepSeek reasoning capabilities ollama run deepseek-r1:7b "Solve this problem: If a train travels 120km in 2 hours, what is its speed?" Enter fullscreen mode Exit fullscreen mode Test via OpenClaw # Send a test message through OpenClaw openclaw chat "Hello, what's 2 + 2?" Enter fullscreen mode Exit fullscreen mode Test Platform Integrations Once your platforms are configured:

Check OpenClaw logs to see what's happening:

# View recent logs openclaw logs --recent # View live logs openclaw logs --follow Enter fullscreen mode Exit fullscreen mode Advanced Setup Tips GPU Acceleration If you have an NVIDIA GPU, enable CUDA acceleration:

# Verify GPU is detected ollama list # Run with GPU acceleration (automatic if GPU available) ollama run deepseek-r1:7b --gpu Enter fullscreen mode Exit fullscreen mode Create Custom Models Use system prompts to create specialized versions:

# Create a Modelfile echo 'FROM deepseek-r1:7b SYSTEM """You are a Python expert. Provide clean, PEP 8 compliant code. """' > /tmp/python-expert # Create the custom model ollama create python-expert -f /tmp/python-expert # Use it in OpenClaw openclaw models set ollama/python-expert Enter fullscreen mode Exit fullscreen mode Multi-Model Setup Run different models for different tasks:

# In config.yaml - configure multiple model presets models: default: ollama/deepseek-r1:7b coding: ollama/deepseek-coder:7b reasoning: ollama/deepseek-r1:14b Enter fullscreen mode Exit fullscreen mode Then switch between them:

# Use coding model openclaw models set coding # Use reasoning model for complex tasks openclaw models set reasoning Enter fullscreen mode Exit fullscreen mode Performance Optimization For better performance:

# Check current model and resources openclaw status --verbose # Monitor Ollama directly ollama list Enter fullscreen mode Exit fullscreen mode Troubleshooting Common Issues Model Won't Load (Out of Memory) Problem: Ollama fails to load the model due to insufficient RAM.

Problem: OpenClaw reports connection errors to Ollama.

Problem: Can't connect Telegram/Discord/WhatsApp.

Yes, DeepSeek is open-source and free to run locally. You only need to provide the hardware (computer with RAM). No API fees, no subscriptions.

Yes, both DeepSeek and OpenClaw have permissive licenses that allow commercial use. Always review the latest license terms.

DeepSeek can run on CPU-only systems. Expect slower inference (a few seconds per response instead of milliseconds). The smaller models (1.5B-7B) work reasonably well on CPU.

Yes, but each model requires additional RAM. A typical setup might run the 7B model alongside a smaller specialist model for specific tasks.

ollama pull deepseek-r1:7b Enter fullscreen mode Exit fullscreen mode Ollama will automatically update if a newer version is available.

Yes, OpenClaw provides API endpoints and webhooks for custom integrations. Check the OpenClaw documentation for details.

If you'd rather avoid the DeepSeek API and keep everything on your own hardware, running OpenClaw backed by Ollama gives you the same interface with any locally hosted model you already have.

You now have a powerful, free AI assistant running locally on your machine. DeepSeek provides the intelligence, OpenClaw provides the agency, and Ollama makes it all work seamlessly.

The combination of DeepSeek and OpenClaw delivers capabilities that would cost hundreds of dollars per month with cloud alternatives — all running on hardware you own.

Ready to build professional AI applications? Download Apidog free and test your AI service integrations with a visual interface designed for developers. Try Apidog's API testing suite to ensure your AI workflows are robust and reliable.