Summary

Adds "gemini" as a new web search backend that uses Gemini API's native google_search grounding tool. This provides web search without requiring any third-party API key — it reuses the existing GOOGLE_API_KEY or GEMINI_API_KEY already configured for the LLM provider.

Motivation

Currently, web_search requires a third-party search API key (Firecrawl, Tavily, Exa, or Parallel). Users who choose Gemini as their LLM provider already have a Google API key, but can't use it for web search. Gemini's native Google Search grounding provides high-quality web search results at no additional cost.

How it works

• Calls Gemini native API with tools: [{"google_search": {}}]
• Gemini automatically generates search queries, searches Google, and returns grounded responses with source URLs
• Uses gemini-2.5-flash for search (fast and cost-effective)
• Returns results in the same normalized format as other backends

Changes

• tools/web_tools.py: Add _gemini_search() function, register "gemini" backend in _get_backend() and _is_backend_available()
• hermes_cli/tools_config.py: Add "Google Gemini" as a search provider option in hermes tools setup wizard

Configuration

# config.yaml
web:
  backend: "gemini"

# .env — same key used for the LLM provider
GOOGLE_API_KEY=your_key_here

When web.backend is unset, Gemini is auto-selected if GOOGLE_API_KEY is present (highest priority in the fallback chain since it doesn't need an additional API key).

Test plan

• hermes tools → select "Google Gemini" as web backend
• hermes chat -q "What happened in tech news today?" → should return grounded results
• hermes doctor → web check passes with only GOOGLE_API_KEY set
• Existing backends (Firecrawl, Tavily, etc.) unaffected