How to Build a Free Local AI Assistant with OpenClaw and DeepSeek

TL;DR

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.

Try Apidog today

Introduction

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:

• Ollama serving DeepSeek models on localhost
• OpenClaw connected as the agent layer, reachable from Telegram, Discord, or WhatsApp
• A tested API endpoint (verified with Apidog) you can reuse in your own apps

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.

Why DeepSeek + OpenClaw?

The Power of DeepSeek

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.

The Flexibility of OpenClaw

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

It provides:

• Multi-platform integration: WhatsApp, Telegram, Discord, Slack, and more
• Autonomous actions: Send emails, manage calendars, browse web, execute commands
• Persistent memory: Remembers context across sessions
• Skills ecosystem: 700+ community-built extensions via ClawHub
• Privacy-focused: Runs entirely locally

Why This Combination Works

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

• Zero API costs
• Complete data privacy
• Customizable behavior
• Full control over your AI assistant
• Multi-platform access

Prerequisites

Before starting, make sure you have:

• A computer with sufficient RAM (see the table below)
• Administrator~ access to install software
• An internet connection for the initial downloads
• Basic familiarity with the command line (we'll explain each step)

RAM Requirements by Model

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

Step 1: Install Ollama

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

macOS

# Using Homebrew (recommended) brew install ollama # Or using the install script curl -fsSL https://ollama.ai/install.sh | sh 

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 

Windows

Download and run the installer from ollama.ai.

Verify the Installation

ollama --version 

You should see output like ollama version 0.5.0 or similar.

Start the Ollama Service

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 

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 

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 

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 

Run the Model

# Interactive chat mode ollama run deepseek-r1:7b 

Type your message and press Enter. Type /exit to quit.

Test with Python

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"]) 

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.

1. Create a new Request in Apidog
2. Set the method to POST
3. Enter the URL: http://localhost:11434/api/generate
4. Add a Header: Content-Type: application/json

Add a Body (JSON):

{ "model": "deepseek-r1:7b", "prompt": "Hello, world!", "stream": false } 

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.

Use the Ollama Python Library

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"]) 

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 

Initial Setup

Run OpenClaw for the first time:

npx openclaw 

This will guide you through initial configuration:

• Set up your first platform connection (Telegram, Discord, etc.)
• Configure basic preferences
• Start the assistant

Verify OpenClaw Is Running

# Check OpenClaw status openclaw status 

Step 4: Connect DeepSeek to OpenClaw

Now the magic happens — we connect DeepSeek as the brain of your OpenClaw assistant.

Method 1: Use Ollama as Backend

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 

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 

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 

Test the Integration

# Test that OpenClaw is using DeepSeek openclaw models status 

You should see output confirming DeepSeek-R1 is active.

Chat Through Your Platform

Now you can chat with DeepSeek through any connected platform:

• Telegram: Send a message to your OpenClaw bot on Telegram.
• Discord: Mention your OpenClaw bot in Discord.
• WhatsApp: Message your OpenClaw WhatsApp number.

The response will come from DeepSeek running locally!

Configuration and Optimization

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

Temperature and Top-P

Control response creativity:

# In config.yaml ollama: temperature: 0.7 # 0.0 = focused, 1.0 = creative top_p: 0.9 # Nucleus sampling top_k: 40 # Token selection 

Context Length

Adjust for longer conversations:

ollama: context_size: 4096 # Increase for longer context 

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. 

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 

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?" 

Test via OpenClaw

# Send a test message through OpenClaw openclaw chat "Hello, what's 2 + 2?" 

Test Platform Integrations

Once your platforms are configured:

• Telegram: Send /start to your OpenClaw bot.
• Discord: Mention your bot with @your-bot hello.
• WhatsApp: Send a message to your configured WhatsApp number.

Monitor Logs

Check OpenClaw logs to see what's happening:

# View recent logs openclaw logs --recent # View live logs openclaw logs --follow 

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 

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 

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 

Then switch between them:

# Use coding model openclaw models set coding # Use reasoning model for complex tasks openclaw models set reasoning 

Performance Optimization

For better performance:

• Close unnecessary applications to free RAM
• Use the smallest model that meets your needs
• Consider upgrading RAM if you frequently hit limits
• Use SSD storage for faster model loading

Monitor Resource Usage

# Check current model and resources openclaw status --verbose # Monitor Ollama directly ollama list 

Troubleshooting Common Issues

Model Won't Load (Out of Memory)

Problem: Ollama fails to load the model due to insufficient RAM.

Solutions:

• Use a smaller model (7B instead of 14B)
• Close other applications to free RAM
• Add more RAM to your system

Slow Responses

Problem: Responses take too long.

Solutions:

• Use a smaller model
• Enable GPU acceleration
• Reduce context size
• Use a faster storage drive (SSD)

OpenClaw Can't Connect to Ollama

Problem: OpenClaw reports connection errors to Ollama.

Solutions:

• Verify Ollama is running: ollama serve
• Check the host in config (default: http://localhost:11434)
• Restart Ollama: pkill ollama && ollama serve

Platform Connection Issues

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

Solutions:

• Verify your API credentials are correct
• Check the platform's API status
• Review OpenClaw logs for specific error messages

FAQ

Is DeepSeek really free to use?

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.

Can I use DeepSeek commercially with OpenClaw?

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

What if I don't have a GPU?

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.

How do I choose between DeepSeek-R1 and DeepSeek-V3?

• DeepSeek-R1: Best for reasoning tasks, math, coding, and problem-solving
• DeepSeek-V3: Best for general-purpose conversation and tasks

Can I run multiple DeepSeek models at once?

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

How do I update DeepSeek to the latest version?

ollama pull deepseek-r1:7b 

Ollama will automatically update if a newer version is available.

Can I connect OpenClaw to my own applications?

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.

Conclusion

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.

What you can do now:

• Chat with DeepSeek through Telegram, Discord, WhatsApp, or other platforms
• Automate tasks like sending emails and managing calendars
• Build custom AI workflows with full privacy
• Scale from the smallest model to the most powerful as your needs grow

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.

Next steps:

• Experiment with different DeepSeek model sizes
• Explore OpenClaw's skill marketplace (ClawHub)
• Connect additional platforms to your assistant
• Create custom prompts for specific use cases

The only limit is your imagination.

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.