Feature: Ollama Web Search Integration — API Key Free Search

[naveed949]

Summary

This PR adds Ollama as a new web search provider for Deep Research, enabling users to perform web searches without requiring external API keys (Tavily, SerpApi). Users simply need a local Ollama instance running with a web-search-capable model.

Motivation

Currently, llm_wiki's Deep Research feature requires users to configure a paid API key from Tavily or SerpApi to perform web searches. This creates a barrier for users who:

• Want to experiment with Deep Research without creating accounts
• Have privacy concerns about sending queries to third-party services
• Already have Ollama running locally for LLM inference

By leveraging Ollama's built-in /api/experimental/web_search endpoint, users can get started with web search immediately if they already use Ollama as their LLM provider.

Changes

Type System (src/stores/wiki-store.ts)

• Added "ollama" to the SearchProvider union type
• Added ollamaUrl?: string to SearchApiConfig interface
• Added ollamaUrl?: string to SearchProviderOverride interface
• Updated SearchProviderConfigs to include "ollama" in the record type

Search Implementation (src/lib/web-search.ts)

• New function: ollamaSearch(query, ollamaUrl, maxResults) — calls Ollama's /api/experimental/web_search endpoint
• New function: hasConfiguredSearchProvider() — utility to check if any valid search provider is configured, now includes Ollama in the check
• Updated resolveSearchConfig() — properly resolves ollama-specific URL configuration with sensible defaults
• Updated webSearch() switch — added case "ollama" to route searches to the new implementation
• Uses Tauri's HTTP plugin (getHttpFetch()) for consistent error handling across all providers
• Implements proper error handling for:
  • Connection refused (Ollama not running)
  • HTTP 401 (requires ollama signin authentication)
  • Other HTTP errors with descriptive messages

UI Changes (src/components/settings/sections/web-search-section.tsx)

• Added Ollama as the first provider in the search settings UI (promoted as the recommended free option)
• Ollama card shows:
  • URL field (pre-filled with http://localhost:11434)
  • No API key field (as it's not needed)
  • Helpful hint explaining requirements
• isLocal: true flag to distinguish local providers from cloud providers in the UI

Internationalization (src/i18n/en.json, src/i18n/zh.json)

• Added ollamaHint translation key explaining that:
  • Uses local Ollama's built-in web search
  • Requires Ollama running with a web-search-capable model

How It Works

1. User selects "Ollama (Local)" in Settings → Web Search
2. URL defaults to http://localhost:11434 (matching the Ollama LLM setting)
3. Deep Research calls webSearch() which routes to ollamaSearch()
4. ollamaSearch() POSTs to {ollamaUrl}/api/experimental/web_search with { query, max_results }
5. Ollama returns JSON with results: [{ title, url, content }]
6. Results are normalized to WebSearchResult[] format

Requirements

• Ollama running locally (or at configured URL)
• A model loaded in Ollama that supports web search (some models may not support this feature)
• For authenticated searches: ollama signin in terminal

API Reference

Ollama Web Search Endpoint:

POST /api/experimental/web_search
Content-Type: application/json

{
  "query": "search terms",
  "max_results": 5
}

Response:
{
  "results": [
    {
      "title": "Result Title",
      "url": "https://example.com",
      "content": "Snippet text..."
    }
  ]
}

Testing

Added new test cases in src/lib/web-search.test.ts:

• calls Ollama web search and normalizes results — verifies correct API call and response parsing
• surfaces Ollama auth errors — verifies 401 handling with actionable message

Screenshots

The new UI shows Ollama as the first (recommended) option:

┌─────────────────────────────────────────────────────┐
│ ┌─────────────────────────────────────────────────┐ │
│ │ ○ Ollama (Local)                      [Active] │ │
│ │ Free web search via your local Ollama instance │ │
│ └─────────────────────────────────────────────────┘ │
│                                                     │
│ ┌─────────────────────────────────────────────────┐ │
│ │ ○ Tavily                                       │ │
│ │ General web search for Deep Research           │ │
│ └─────────────────────────────────────────────────┘ │
│                                                     │
│ ┌─────────────────────────────────────────────────┐ │
│ │ ○ SerpApi                                      │ │
│ │ Google, Bing, DuckDuckGo, Scholar...           │ │
│ └─────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────┘

Comparison with Existing Providers

Provider	API Key Required	URL Config	Notes
Ollama	No	Yes (default: localhost:11434)	Uses local Ollama instance
Tavily	Yes	No	Cloud service
SerpApi	Yes	No	Cloud service
SearXNG	No	Yes	Self-hosted metasearch

Migration Notes

None — this is purely additive. Existing Tavily/SerpApi/SearXNG configurations continue to work unchanged.

Future Enhancements (Out of Scope)

• webFetch integration for fetching full page content
• Model capability detection (which models support web search)
• Auto-detection of Ollama availability on startup