Ollama Web Search — The Complete Guide

A production-tested guide to adding web search to local LLMs via Ollama. Every example was tested on a real Debian 13 server. No fluff.

Author: Martin Simons
Tested on: Ollama 0.18.2, Debian 13 Trixie, Python 3.11, Node.js 22
Date: March 2026

---

Why This Guide Exists

Most tutorials about Ollama web search are wrong in at least one of these ways:

• They use DuckDuckGo's free API — which returns empty results
• They don't handle streaming — leading to 60-second timeouts
• They don't document which models actually support tool calling
• They skip production concerns like auto-restart, error handling, and privacy

This guide fixes all of that. Every code example was run on a real server.

---

The Hard Truth About DuckDuckGo's Free API

The most commonly recommended approach — DuckDuckGo's free API — simply does not work:

curl "https://api.duckduckgo.com/?q=ollama+latest+version&format=json&no_html=1"

Result:

{
  "Abstract": "",
  "AbstractText": "",
  "RelatedTopics": [],
  "Results": []
}

Empty. Every time. For almost every query.
Do not use DuckDuckGo's free API for Ollama web search.

---

Search Backend Comparison

Backend	API Key	Real Results	Privacy	Cost	Verdict
DuckDuckGo free API	❌ None	❌ Empty	✅	Free	❌ Don't use
SearXNG self-hosted	❌ None	✅ Real	✅ Full	Free	✅ Best overall
Google Custom Search	✅ Required	✅ Excellent	❌ Cloud	100/day free	✅ Best results
Bing Search API	✅ Required	✅ Excellent	❌ Cloud	1000/month free	✅ Good alternative
OpenClaw built-in	❌ None	✅ Real	⚠️ Depends	Free	⚠️ GPU recommended

---

Model Comparison for Tool Calling

Tested on Ollama 0.18.2, CPU-only server, 32GB RAM:

Model	Tool Calling	Thinking Mode	Web Search Speed	Recommended
qwen3:8B	✅ Excellent	On by default	16+ min on CPU	⚠️ GPU only
llama3.1:8B	✅ Good	None	2-5 min on CPU	✅ CPU servers
mistral-nemo:12b	⚠️ Moderate	None	3-7 min on CPU	⚠️ Sometimes
deepseek-r1:14b	⚠️ Limited	Always on	Very slow	❌ Not for search

⚠️ qwen3 thinking mode warning: qwen3 enables extended reasoning by default. On CPU-only servers this causes 16+ minute response times for web search queries. A GPU with 16GB+ VRAM is strongly recommended for qwen3.

---

Prerequisites

System Requirements

• Ollama 0.17.1+ (tested on 0.18.2)
• Python 3.10+ or Node.js 22+
• Docker (for SearXNG)
• 16GB+ RAM recommended

Install Ollama

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

Pull a recommended model

ollama pull llama3.1        # Best for CPU-only servers
ollama pull qwen3           # Best quality (GPU recommended)

Set up Python environment

mkdir -p /opt/projects/ollama-websearch
cd /opt/projects/ollama-websearch
python3 -m venv .venv
source .venv/bin/activate
pip install requests ollama

Install Node.js 22 (via nvm)

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash
source ~/.bashrc
nvm install 22
nvm use 22
nvm alias default 22

---

Option 1: SearXNG (Recommended — Private, Free, No API Key)

Setup SearXNG with Docker

# Create config directory
mkdir -p /opt/projects/ollama-websearch/docker/searxng

# Start SearXNG
docker run -d \
  --name searxng \
  --restart always \
  -p 8080:8080 \
  -e BASE_URL=http://localhost:8080 \
  searxng/searxng:latest

Enable JSON format (required)

# Copy default config
docker cp searxng:/etc/searxng/settings.yml \
  /opt/projects/ollama-websearch/docker/searxng/settings.yml

Edit settings.yml — find the search: section and add:

search:
  safe_search: 0
  autocomplete: ""
  formats:
    - html
    - json

server:
  secret_key: "change-this-to-a-random-string"
  limiter: false

Restart with config mounted:

docker stop searxng && docker rm searxng

docker run -d \
  --name searxng \
  --restart always \
  -p 8080:8080 \
  -e BASE_URL=http://localhost:8080 \
  -v /opt/projects/ollama-websearch/docker/searxng/settings.yml:/etc/searxng/settings.yml \
  searxng/searxng:latest

Verify it works:

curl "http://localhost:8080/search?q=ollama&format=json" | python3 -m json.tool | head -20

Python Example

source .venv/bin/activate
./websearch_searxng.py "What is the latest version of Ollama?"

Expected output:

❓ Question: What is the latest version of Ollama?
🤔 Thinking...
🔍 Searching SearXNG for: Ollama latest version
📄 Found 5 results
💬 Answer:
The latest version of Ollama is 0.18.2.

Node.js Example

node websearch_searxng.js "What is the latest version of Ollama?"

Shell/curl Example

./websearch_searxng.sh "What is the latest version of Ollama?"

---

Option 2: Google Custom Search API

Setup

1. Go to Google Cloud Console
2. Enable Custom Search API
3. Create API key at Credentials
4. Create search engine at Programmable Search Engine

export GOOGLE_API_KEY=your-api-key
export GOOGLE_CSE_ID=your-search-engine-id

# Python
./websearch_google_bing.py "What is the latest version of Ollama?"

# Node.js
node websearch_google_bing.js "What is the latest version of Ollama?"

# Shell
./websearch_google_bing.sh "What is the latest version of Ollama?"

---

Option 3: Bing Search API

Setup

1. Go to Azure Portal
2. Create Bing Search v7 resource (free tier: 1000 queries/month)
3. Copy API key

export BING_API_KEY=your-api-key
export SEARCH_BACKEND=bing

# Python
./websearch_google_bing.py "What is the latest version of Ollama?"

# Node.js
node websearch_google_bing.js "What is the latest version of Ollama?"

# Shell
./websearch_google_bing.sh "What is the latest version of Ollama?"

---

Option 4: OpenClaw Built-in Web Search

Ollama 0.18+ ships with OpenClaw integration which includes built-in web search.

Requirements

• Node.js 22.12+ (use nvm)
• npm installed under nvm (not system npm)
• psmisc package (sudo apt install psmisc)
• GPU recommended (CPU-only causes runaway inference)

Setup

# Install psmisc (required for --force flag)
sudo apt install psmisc -y

# Install Node.js 22 via nvm
nvm install 22 && nvm use 22

# Install openclaw under nvm
/home/$USER/.nvm/versions/node/v22.22.1/bin/npm install -g openclaw

Run as systemd service (recommended)

sudo cat > /etc/systemd/system/openclaw-gateway.service << 'SYSTEMD'
[Unit]
Description=OpenClaw Gateway
After=network.target ollama.service
Wants=ollama.service

[Service]
Type=simple
User=YOUR_USERNAME
Environment="PATH=/home/YOUR_USERNAME/.nvm/versions/node/v22.22.1/bin:/usr/local/bin:/usr/bin:/bin"
ExecStart=/home/YOUR_USERNAME/.nvm/versions/node/v22.22.1/bin/openclaw gateway
Restart=always
RestartSec=5

[Install]
WantedBy=multi-user.target
SYSTEMD

sudo systemctl daemon-reload
sudo systemctl enable openclaw-gateway
sudo systemctl start openclaw-gateway

Then launch:

ollama launch openclaw --model llama3.1:latest

⚠️ Known issues on CPU-only servers:

• qwen3 thinking mode causes 16+ minute hangs — use llama3.1 instead
• Set "restart": false in ~/.openclaw/openclaw.json to prevent session auto-resume
• Clear workspace after crashes: rm -rf ~/.openclaw/workspace/* /tmp/openclaw

---

Production Tips

Ollama on multi-user / Docker servers

By default Ollama only listens on 127.0.0.1. If you run LiteLLM or other tools in Docker, configure Ollama to listen on all interfaces:

sudo systemctl edit ollama

Add:

[Service]
Environment="OLLAMA_HOST=0.0.0.0:11434"

sudo systemctl daemon-reload && sudo systemctl restart ollama

Choosing the right model

# Fast, reliable tool calling on CPU
ollama pull llama3.1

# Best quality, needs GPU for web search
ollama pull qwen3

# Disable thinking mode in qwen3 for interactive use
# In OpenClaw TUI: /think off

Directory structure for multi-user servers

sudo mkdir -p /opt/projects/ollama-websearch
sudo chown youruser:devteam /opt/projects/ollama-websearch
sudo chmod 2775 /opt/projects/ollama-websearch

---

Quick Reference

# Check Ollama is running
curl http://localhost:11434/api/tags

# Check SearXNG is running
curl "http://localhost:8080/search?q=test&format=json" | jq .results[0].title

# Run Python example
source .venv/bin/activate && ./websearch_searxng.py "your question"

# Run Node.js example
node websearch_searxng.js "your question"

# Run shell example
./websearch_searxng.sh "your question"

---

License

MIT — use freely, attribution appreciated.