Local AI Setup Guide

What You're Building

By the end of this guide, you'll have:

• OpenClaw — A local AI assistant that runs on your computer
• Ollama — Software that runs large language models locally
• Llama 3.1 — A powerful, free, open-source AI model (70B or 405B)
• LawTasksAI — Integrated and ready to process confidential documents privately

Time required: 30-60 minutes (depending on download speeds)

Cost: $0 (all software is free and open-source)

Hardware Requirements

Model	RAM Required	Performance	Recommended For
Llama 3.1 8B	8GB	Basic	Simple tasks, testing
Llama 3.1 70B	32GB+	Excellent	Most legal work
Llama 3.1 405B	64GB+	Best-in-class	Complex analysis, large documents

How to check your RAM:

• Windows: Press Win + Pause/Break or right-click "This PC" → Properties
• Mac: Apple menu → About This Mac
• Linux: Run free -h in terminal

Step-by-Step Setup

OpenClaw is a local AI assistant that connects to your local AI models. It's like having Claude Desktop or ChatGPT, but everything runs on your computer.

For Windows:

1. Go to openclaw.ai
2. Click "Download for Windows"
3. Run the installer
4. Follow the setup wizard

For Mac:

1. Go to openclaw.ai
2. Click "Download for Mac"
3. Open the .dmg file and drag OpenClaw to Applications
4. Launch OpenClaw (you may need to allow it in System Preferences → Security)

For Linux:

# Install via npm (requires Node.js 18+)
npm install -g openclaw

# Or download the .deb/.rpm from openclaw.ai

Verify installation: Open OpenClaw. You should see a chat interface. Type /status to confirm it's running.

Ollama is the software that runs AI models on your computer. Think of it like a local AI engine.

For Windows & Mac:

1. Go to ollama.com
2. Click "Download"
3. Run the installer
4. Ollama will start automatically as a background service

For Linux:

curl -fsSL https://ollama.com/install.sh | sh

Verify installation:

• Windows/Mac: You should see an Ollama icon in your system tray
• All platforms: Open terminal/command prompt and run: ollama --version

Now you'll download the actual AI "brain" that will process your legal documents and research queries. We recommend Llama 3.1 — it's free, powerful, and designed for professional use.

Open your terminal/command prompt and run:

For 32GB+ RAM (Recommended):

ollama pull llama3.1:70b

For 64GB+ RAM (Best Performance):

ollama pull llama3.1:405b

For 8-16GB RAM (Testing Only):

ollama pull llama3.1:8b

You'll see a progress bar. Go get coffee. ☕

Verify the model works:

ollama run llama3.1:70b

# You should see a chat prompt. Type a test question:
>>> What is attorney-client privilege?

# Press Ctrl+D to exit when done

Now we tell OpenClaw to use the local AI model instead of sending data to the cloud.

In OpenClaw:

1. Type /config
2. Look for the model setting
3. Set it to: ollama/llama3.1:70b (or whichever model you downloaded)

Or edit the config file directly:

• Windows: %USERPROFILE%\.openclaw\config.yml
• Mac/Linux: ~/.openclaw/config.yml

Add or update this section:

model: ollama/llama3.1:70b
providers:
  ollama:
    baseUrl: http://localhost:11434

Restart OpenClaw for changes to take effect.

Test it: In OpenClaw, type: What is the statute of limitations for fraud in California?

The response should come from your local model (you may notice it's slightly slower than cloud AI, but everything is private).

Now you're ready to add LawTasksAI skills to your local setup.

In OpenClaw, type:

/skill install lawtasksai

You'll be prompted for your license key (starts with lt_). Enter it.

Don't have a license yet? Purchase credits at lawtasksai.com. You'll receive your license key via email instantly.

Verify installation:

/skill list

You should see lawtasksai in the list.

Let's make sure everything works and nothing is going to the cloud.

Test 1: Check Your Connection

In OpenClaw, type: /status

Look for the model line. It should show ollama/llama3.1:70b (or your chosen model).

Test 2: Disconnect Your Internet

1. Turn off Wi-Fi or unplug ethernet
2. In OpenClaw, ask: Summarize the attorney-client privilege doctrine
3. You should still get a response (proving it's running locally)
4. Reconnect to the internet

Test 3: Run a LawTasksAI Document Task

In OpenClaw, type: What legal document tasks do you have?

The AI should list available tasks. Try one that processes documents locally, like:

• Analyze this deposition transcript (attach a PDF)
• Summarize this contract (attach a Word doc)
• Review this discovery response (attach a document)

The document will be processed entirely on your machine. LawTasksAI will only send a license verification request (not the document contents).

Troubleshooting

Problem: "Model not found" or "Connection refused"

Solution:

• Make sure Ollama is running (check system tray for Ollama icon)
• Verify the model downloaded correctly: ollama list
• Check that OpenClaw config points to http://localhost:11434
• Restart Ollama service

Problem: Responses are very slow

Solution:

• You may not have enough RAM. Check system monitor while running.
• Try a smaller model: ollama pull llama3.1:8b
• Close other applications to free up memory
• Consider upgrading your hardware or using a smaller model for routine tasks

Problem: "License key invalid" when using LawTasksAI

Solution:

• You need an internet connection for license verification (just not for AI processing)
• Check that your license key starts with lt_
• Verify your key at lawtasksai.com
• Email support@lawtasksai.com if issues persist

Problem: OpenClaw crashes or freezes

Solution:

• Your model may be too large for your RAM. Use llama3.1:8b instead.
• Check logs: /logs in OpenClaw or check ~/.openclaw/logs/
• Restart OpenClaw and try a simpler prompt first

Performance Tips

Speed Up Responses

• Use GPU acceleration if you have a compatible NVIDIA GPU (requires CUDA setup)
• Quantized models: Smaller, faster models that sacrifice a bit of quality: ollama pull llama3.1:70b-q4
• Adjust context window: Smaller context = faster responses but less memory

Save Disk Space

Models are stored in:

• Windows: C:\Users\YourName\.ollama\models
• Mac: ~/.ollama/models
• Linux: ~/.ollama/models

Delete unused models: ollama rm llama3.1:8b

Batch Processing

For large document review projects, consider:

• Processing documents overnight (schedule with OpenClaw cron)
• Breaking large PDFs into smaller chunks
• Using the 405B model for the most critical analysis, 70B for routine work

Comparing Cloud vs. Local

Feature	Cloud AI (Claude/ChatGPT)	Local AI (This Setup)
Privacy	⚠️ Data sent to third-party	✅ Everything stays local
Speed	✅ Very fast	⚠️ Depends on your hardware
Cost	$20-60/month	$0/month (after hardware)
Setup	✅ 5 minutes	⚠️ 30-60 minutes
Hardware Required	✅ Any computer	⚠️ 32GB+ RAM recommended
Rule 1.6 Compliance	⚠️ Requires client consent	✅ Fully compliant
Internet Required	✅ Yes (always)	⚠️ Only for license check
Quality	✅ Best-in-class	✅ Excellent (with 70B+)

Upgrading to GPU Acceleration (Advanced)

If you have an NVIDIA GPU, you can dramatically speed up local AI processing:

1. Install NVIDIA CUDA Toolkit: developer.nvidia.com/cuda-downloads
2. Verify GPU is detected: nvidia-smi
3. Ollama will automatically use GPU if available
4. Check GPU usage while running: nvidia-smi in a separate terminal

Performance boost: 5-10x faster responses with a modern GPU.

Need Help?

If you get stuck:

• Email us: support@lawtasksai.com — we'll walk you through it
• OpenClaw docs: docs.openclaw.ai
• Ollama docs: github.com/ollama/ollama

What's Next?

Now that you have a fully private setup:

• Explore the task catalog: lawtasksai.com/skills
• Read the security documentation: lawtasksai.com/security
• Create a usage policy: Document when to use local vs. cloud AI
• Train your team: Show paralegals and associates how to use the system

Back to LawTasksAI Home Security Overview