Web Search Integration

Enable AI-powered web search with your local Ollama models.

Which Setup Should I Use?

Both setups use SearXNG for multi-engine search - you always get the full experience:

Setup	Interface	Best For
Open WebUI	ChatGPT-like chat	General chat with web search capability
Perplexica	Perplexity-like research	AI research with inline citations

What is SearXNG?

SearXNG is a meta-search engine - it queries multiple search engines at once and combines their results:

┌─────────────┐
│   SearXNG   │──→ DuckDuckGo ──→ results
│  (combines  │──→ Google ──────→ results  ──→ Aggregated
│   results)  │──→ Bing ────────→ results      Results
│             │──→ Brave ───────→ results
│             │──→ Wikipedia ───→ results
└─────────────┘

Quick Decision Guide

Want This?	Install This	Containers
Chat + multi-engine web search	Open WebUI	2 (Open WebUI + SearXNG)
AI research with citations (Perplexity-like)	Perplexica	3 (Perplexica + SearXNG)
Everything	Both	4 (share SearXNG)

note

Open WebUI now always includes SearXNG for multi-engine search. There's no "basic search" option - you get the full experience by default.

Key Insight

If you install Perplexica, you get SearXNG automatically. Open WebUI can then use that same SearXNG instance for multi-engine search too!

Why Web Search?

LLMs have a knowledge cutoff date. Web search integration allows your local AI to:

• Answer questions about current events
• Look up documentation and APIs
• Research topics in real-time
• Cite sources for factual claims

Feature Comparison

Feature	Open WebUI	Perplexica
Interface	ChatGPT-like	Perplexity-like
Search engines	Multi-engine via SearXNG	Multi-engine via SearXNG
Privacy	100% self-hosted	100% self-hosted
Setup complexity	2 containers	3 containers
Best for	General chat	Privacy-focused research
Citations	Basic source links	Numbered references throughout

Option 1: Open WebUI

A beautiful, feature-rich chat interface with built-in web search.

Installation

Recommended (single-user mode - no login required):

.\setup-ollama-websearch.ps1 -Setup OpenWebUI -SingleUser

Standard (with user accounts):

.\setup-ollama-websearch.ps1 -Setup OpenWebUI

Access

Open http://localhost:3000

What Gets Pre-Configured

The setup script automatically configures:

• Web search - Pre-enabled with DuckDuckGo (or SearXNG if available)
• Models - All your installed Ollama models pre-selected
• Session persistence - Secret key saved so you stay logged in across restarts

First-Time Setup

With -SingleUser: Nothing to do! Open the URL and start chatting.

Without -SingleUser:

1. Create an account (first user becomes admin)
2. Web search is pre-configured automatically

Web Search Configuration

Open WebUI always uses SearXNG for web search, which is automatically started alongside it:

Component	Port	Description
Open WebUI	3000	Chat interface
SearXNG	4000	Multi-engine search aggregator

This gives you:

• No rate limits - Self-hosted search
• Multi-engine - Queries DuckDuckGo, Google, Bing, Brave, etc. simultaneously
• Complete privacy - All searches stay on your network

To customize search engines, edit searxng/settings.yml after installation.

Usage

1. Click the + button next to the message input
2. Toggle Web Search on
3. Type your query - results will include web sources

Or prefix your query with /web:

/web What's the latest version of Node.js?

Why Open WebUI?

• Familiar interface - If you've used ChatGPT, you'll feel at home
• Rich features - File uploads, image generation, code execution
• Active development - Frequent updates, large community
• Easy setup - One container, minimal configuration

Option 2: Perplexica + SearXNG

A Perplexity AI alternative that's 100% self-hosted and private.

Installation

.\setup-ollama-websearch.ps1 -Setup Perplexica

Access

• Perplexica: http://localhost:3002
• SearXNG: http://localhost:4000

Architecture

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│  Perplexica     │────▶│    SearXNG      │────▶│  Search Engines │
│  Frontend       │     │  (Meta-Search)  │     │  (DDG, Google)  │
│  :3002          │     │  :4000          │     │                 │
└────────┬────────┘     └─────────────────┘     └─────────────────┘
         │
         ▼
┌─────────────────┐
│  Perplexica     │────▶ Ollama (:11434)
│  Backend        │
│  :3001          │
└─────────────────┘

Why Perplexica?

• Complete privacy - No data leaves your network
• Source citations - Every answer includes references
• Focus modes - Academic, writing, Wolfram Alpha, YouTube, Reddit
• Meta-search - SearXNG aggregates multiple search engines

First-Time Setup

1. Open http://localhost:3002
2. Click the settings icon (gear) in the sidebar
3. Under Chat Model, select Ollama and choose a model:
   • qwen2.5:3b for fast queries
   • qwen2.5:14b for better synthesis
4. Under Embedding Model, select Local and choose BGE Small
5. Click Save and start searching

Using Perplexica

Type your question and press Enter. Perplexica will:

1. Search the web via SearXNG
2. Read and analyze relevant sources
3. Synthesize an answer with citations

Focus Modes

Select a focus mode before searching for optimized results:

Mode	Best For
All	General web searches
Academic	Research papers and scholarly articles
YouTube	Finding video content
Reddit	Community discussions and opinions
Wolfram Alpha	Math, calculations, data queries
Writing	Writing help (no web search)

Tips

• Be specific: "React 19 new features 2024" works better than "tell me about React"
• Use Academic mode for technical documentation
• Larger models (32B) give better synthesis but are slower

Configuration

The setup script auto-generates perplexica/config.toml with the correct Ollama URL:

• Docker: Uses host.docker.internal:11434
• Podman: Auto-detects gateway IP (e.g., 172.x.x.1:11434)

To manually configure, edit perplexica/config.toml:

[GENERAL]
PORT = 3001
SIMILARITY_MEASURE = "cosine"
KEEP_ALIVE = "5m"

[MODELS.OLLAMA]
API_URL = "http://172.17.144.1:11434"

[MODELS.OPENAI]
API_KEY = ""

[API_ENDPOINTS]
SEARXNG = "http://searxng:8080"

Container Config Path

Inside the container, the config is at /home/perplexica/config.toml (not /app/).

SearXNG Customization

Edit searxng/settings.yml to enable/disable search engines:

engines:
  - name: duckduckgo
    disabled: false
  - name: google
    disabled: false  # Requires no API key
  - name: wikipedia
    disabled: false
  - name: github
    disabled: false
  - name: stackoverflow
    disabled: false

International Search Engines

Yandex Cloud API Setup

The built-in Yandex engine in SearXNG is blocked by CAPTCHAs. Use Yandex Cloud's official Search API instead for reliable Russian-language search.

Step 1: Create Yandex Cloud Account

1. Go to console.yandex.cloud
2. Sign up or log in with your Yandex account
3. Create a billing account (free tier available)

Step 2: Create Service Account

1. In the Yandex Cloud console, go to your folder
2. Navigate to IAM → Service accounts
3. Click Create service account
4. Name it searxng-search
5. Click Create

Step 3: Assign Search API Role

1. Click on your new service account
2. Go to Roles tab
3. Click Assign role
4. Add role: search-api.webSearch.user

Step 4: Generate API Key

1. In your service account, go to API keys tab
2. Click Create API key
3. Select scope: yc.search-api.execute
4. Save the key immediately - shown only once!

Step 5: Configure Environment

Create a .env file in the project root (never commit this!):

# Copy from .env.example
cp .env.example .env

# Edit with your credentials
YANDEX_API_KEY=your-api-key-here
YANDEX_FOLDER_ID=your-folder-id-here

Your folder ID is in the console URL: console.yandex.cloud/folders/YOUR_FOLDER_ID

Step 6: Restart SearXNG

# Restart to pick up new environment variables
podman-compose -f docker-compose-perplexica.yml down
podman-compose -f docker-compose-perplexica.yml up -d

Usage

Yandex Cloud Search is now available alongside other engines. Your queries will automatically include Russian search results.

Free Tier

Yandex Cloud offers 10,000 free search queries per day - more than enough for personal use.

Baidu Search (Chinese)

For Chinese-language search, Baidu integration requires CAPTCHA handling. See the advanced search configuration section.

Advanced: Multi-Language Search

For comprehensive international search with translation:

1. Query in English → Results from all engines
2. SearXNG aggregates → DuckDuckGo + Google + Yandex + Baidu
3. LLM synthesizes → Translates and combines results

The LLM (GPT-4, Claude, Qwen) naturally handles translation when presenting results to you in English.

Comparing Search Results

Open WebUI Search

User: What's new in Python 3.13?

AI: Based on my web search, Python 3.13 was released on October 7, 2024
with these key features:
- Free-threaded mode (experimental)
- JIT compiler (experimental)
- Improved error messages
- ...

[Sources: python.org, realpython.com]

Perplexica Search

User: What's new in Python 3.13?

AI: # Python 3.13 Release Notes

Python 3.13 introduced several significant changes:

## Free-Threaded Mode
The GIL can now be disabled experimentally... [1]

## JIT Compiler
A new JIT compiler improves performance... [2]

Sources:
[1] docs.python.org/3/whatsnew/3.13.html
[2] realpython.com/python313-new-features/

Best Practices

Model Selection for Search

The setup script installs optimized models for web search (RTX 5090, 32GB VRAM):

Task	Recommended Model	VRAM
Quick lookups	qwen2.5:3b	~4GB
Synthesis & code	qwen2.5-coder:14b	~17GB

Total: ~21GB - both models fit in VRAM simultaneously, leaving ~11GB for context.

For deep research or academic work, you can manually load larger models like qwen3:32b or deepseek-r1:32b (Ollama will swap as needed).

Prompt Engineering

For better search results:

Instead of: "Tell me about React"
Try: "What are the new features in React 19 released in 2024?"

Specific questions get better search results.

Rate Limiting

Some search providers have rate limits. For heavy usage:

• Use SearXNG (self-hosted, no limits)
• Rotate between providers
• Cache frequent queries

Troubleshooting

"No search results"

1. Check search provider is configured
2. Verify internet connectivity from container
3. Try a different search provider

"Container can't reach Ollama"

This is a common Podman/WSL2 issue. See Podman Networking.

"SearXNG returns empty results"

Some engines may be blocked or rate-limited:

# Access SearXNG directly to test
curl http://localhost:4000/search?q=test&format=json

Testing Your Setup

By default, the setup script just starts containers quickly. Use -Test to verify everything works:

.\setup-ollama-websearch.ps1 -Test

What Gets Tested

Test	What It Checks
Container health	All containers running and healthy
Model inference	Each model responds to a simple prompt
Web endpoints	Open WebUI, SearXNG, Perplexica respond

Troubleshooting

If something isn't working, use -Diagnose:

.\setup-ollama-websearch.ps1 -Diagnose

Shows: container runtime, Ollama status, container status, network config, and logs for unhealthy containers.

Manual Testing

Test SearXNG engines individually:

.\test-searxng-engines.ps1

Sample output:

[OK]   duckduckgo    (3 results, 0.8s)
[OK]   google        (5 results, 1.2s)
[WARN] bing          (0 results - may be rate-limited)
[OK]   wikipedia     (2 results, 0.5s)

Summary: 3/4 engines working

Test API Directly

# Test model inference
$body = @{model="qwen2.5:3b"; prompt="Say OK"; stream=$false} | ConvertTo-Json
Invoke-RestMethod -Uri "http://localhost:11434/api/generate" -Method Post -Body $body -ContentType "application/json"

# Test SearXNG
Invoke-RestMethod -Uri "http://localhost:4000/search?q=test&format=json"