---
title: "Search"
id: "1734"
type: "page"
slug: "search"
published_at: "2026-09-20T20:56:04+00:00"
modified_at: "2026-11-18T17:59:48+00:00"
url: "https://xedant.com/agents/research/docs/search/"
markdown_url: "https://xedant.com/agents/research/docs/search.md"
excerpt: "Search is the entry point of any research. Research works with eleven providers and can…"
---

# Search

[https://xedant.com/agents/research/docs/search/.md](https://xedant.com/agents/research/docs/search/.md)

Search is the entry point of any research. Research works with eleven providers and can spread queries across them with automatic backup: one service being down does not stop the research.

## Eleven providers

- **Serper** — search over Google, API key;
- **Keenable** — API key;
- **Brave** — API key;
- **Tavily** — API key, also returns a ready-made AI answer;
- **TinyFish** — free search over Google’s live index, API key; free up to 12,000 requests a day and 30 requests a minute per key;
- **Exa** — search by meaning rather than by matching words: pages are picked by the sense of the query, and ready text fragments come back with them, API key;
- **Anysite** — live results from Bing, Brave, Baidu, Startpage or Yandex, collected on demand, API key; the search engine is chosen in the service’s settings. Noticeably slower than the rest: results are gathered live instead of being taken from a ready-made index;
- **DuckDuckGo** — works without a key;
- **SearXNG** — a self-hosted metasearch engine you run yourself, connected by its base URL;
- **JSON SEO** — results from Yandex, Google or Bing, API key; the search engine and the region are set in the service’s settings. The choice of engine matters: Yandex covers the Russian-language results better, Google and Bing the global ones, and the region narrows down the geography of the results.

The eleventh provider can be your own Python script: if the service you need is not among the ready-made ones, it plugs in as a regular service, with a priority and fallbacks (details in [Python Scripts](/agents/research/docs/scripts)
).

## Setup

Providers are configured in the **Services** section of the web interface — one service instance per provider. The API keys for Serper, Keenable, Brave, Tavily, JSON SEO, TinyFish, Exa and Anysite are set right there; JSON SEO and Anysite additionally pick a search engine (for JSON SEO — Yandex, Google or Bing; for Anysite — Bing, Brave, Baidu, Startpage or Yandex), and JSON SEO also, optionally, a region. For SearXNG — the base address of your own instance. There are no environment variables for the keys. Every service can be tested right from the same screen, and each gets a priority for fallbacks. For details, see [Services & Models](/agents/research/docs/services)
.

## Search sessions

Search runs in sessions: queries go out in a batch, statuses update in real time. The Searches section shows the list of sessions with filtering and pagination, the statuses (Pending, Running, Completed, Failed) and each session’s results. Results can be filtered for relevance before downloading — a position in the results does not mean usefulness, and links are picked by title and snippet.

## Fallback

Queries are distributed across providers: on an error or an outage, the system automatically moves on to the next service by priority. One topic can happily search on several providers at once — for wider coverage of the results.

## Access from external agents

Search is also available from the outside — for example, from another project’s Agent: `POST /api/agent/search` with the `X-API-Key` header (the `RESEARCH_APP_HASH` variable). Without the key the API returns 503. For details, see [Agent Integration](/agents/research/docs/agent)
.

[← Back to documentation](/agents/research/docs)
