Search API Reference

Service retired

The previous design routed search through a standalone Go microservice backed by SearXNG (meta-searching Google/Bing/DuckDuckGo through a Tor SOCKS proxy) with Brave as a fallback. That entire stack — the service, the SearXNG container, and the Tor container — has been decommissioned.

What replaced it

Search is now an internal module of the Backend API (app/common/search_providers.py). It is called in-process by the backend — there is no HTTP endpoint and no dedicated port.

• Primary provider — Serper.dev: POST https://google.serper.dev/search (web) and POST https://google.serper.dev/news (news). Auth via X-API-KEY: SERPER_API_KEY. Paginates ~10 results/page up to 10 pages (~100 results).
• Fallback provider — Brave Search (called only when Serper errors): GET https://api.search.brave.com/res/v1/web/search (20/page) and GET https://api.search.brave.com/res/v1/news/search (50/page). Auth via X-Subscription-Token: BRAVE_API_KEY. Offset pagination 0–9.

Consumers call two entry points — search_web() and search_news() — which return a unified list[SearchResult]. The fallback is error-triggered only: a successful-but-empty Serper response does not invoke Brave.

from app.common.search_providers import search_web, search_news, age_to_freshness

results = await search_web("earthquake Turkey 2026", freshness=age_to_freshness(event.detected_at), limit=100)
news = await search_news("Cyclone Mocha landfall", limit=50)

Configuration is limited to two environment variables in the Backend API (app/config.py): SERPER_API_KEY and BRAVE_API_KEY. There is no SEARCH_API_URL, SEARXNG_URL, Tor, or residential proxy configuration.

Where to read more

For the full architecture of the current search layer — provider roles, the unified SearchResult model, freshness mapping, and configuration — see the Search Layer service page.