Providers & modes.
Name one or several in providers, or leave it out and the router tries them in the order below. The manifest lists the ones serving right now.
The engines#
| provider | good for | modes |
|---|---|---|
| serper | Google results: web, news, Scholar, patents, images and places. First for most text modes. | general news academic deep scholar patents images places |
| exa | Neural search over its own index. Strong on papers, named people, pages similar to a given URL, and deep searches. | general academic people deep similar |
| brave | Brave's independent web index. In deep mode it adds Brave's LLM Context, extracts from the top pages. | general news deep |
| linkup | A results engine: ranked web results with snippets, at standard or deep depth. It does not write answers. | general news deep |
| parallel | Search built for agents, with a fast tier for general and news and an advanced tier for deep. | general news deep |
| tavily | Search built for agents. Basic depth for general and news, advanced for academic and deep. | general news academic deep |
| perplexity | Perplexity's Search API: ranked results from its own index, with allow or deny domain lists. | general news academic deep |
| jina | Jina search for general results, and Jina Reader, which turns a page or a PDF into markdown. | general extract |
| firecrawl | Web search, and page rendering for extract when a page needs a browser. Reads PDFs up to 20 pages. | general extract |
| serpapi | Google News, Scholar and Images through SerpApi. A fallback for those three modes. | news scholar images |
| browserless | A hosted headless browser that renders the page before reading it. Last in the extract chain. | extract |
| xai | Grok's X search: live posts from X, limited by handle and date. Also one of the deep engines. | social deep |
| direct | Our own fetch: a plain request with an honest user agent that asks for markdown, then converts HTML to markdown. No browser and no impersonation. | extract |
| hackernews | Hacker News stories with points, comment counts and authors, for what people say about a topic. In extract, a news.ycombinator.com/item URL returns the whole thread, every comment in order with its author and depth. | community extract |
| Reddit posts with scores, comment counts and authors. With r/<subreddit> in the query it searches that subreddit (cheapest); without one, all of Reddit. In extract, a Reddit thread URL returns the post and every comment in order with its author, score and depth. | community extract | |
| tiktok | TikTok videos with views, likes, comments, shares and author. In extract, a TikTok video URL returns the caption, stats and transcript. | social extract |
| Instagram reels with plays, likes, comments and author. In extract, a post or reel URL returns the caption, stats and top comments. | social extract | |
| youtube | YouTube videos and shorts with views, duration and channel. In extract, a video URL returns the description, stats and full transcript. | social extract |
| One public Facebook post, video or reel by URL: text, reactions, comments, shares, views and a transcript for short videos. Extract only. | extract | |
| threads | One Threads post by URL: text, likes, reposts and replies. Extract only. | extract |
stealth is accepted as the old name of direct. direct is first for extract because it is the cheapest way to read a page that serves its content to a plain request. Pages it cannot read, such as those that need JavaScript or answer with a bot challenge, are rejected and passed to the rendering services. Rejected attempts are not charged.
Router order#
Without providers, the router tries engines in this order, healthy engines first and skipping any that cannot honour your filters. Several modes have one engine: if it is unavailable, the mode descriptions on Search say what to use instead.
| mode | engines, in order |
|---|---|
| general | serper, exa, jina, brave, linkup, parallel, tavily, perplexity, firecrawl |
| news | serper, brave, linkup, parallel, tavily, perplexity, serpapi |
| academic | serper, exa, tavily, perplexity |
| scholar | serper, serpapi |
| people | exa |
| social | xai, tiktok, instagram, youtube |
| community | reddit, hackernews |
| images | serper, serpapi |
| places | serper |
| patents | serper |
| deep | exa, serper, brave, linkup, parallel, tavily, perplexity, xai |
| extract | reddit, hackernews, tiktok, instagram, youtube, facebook, threads, direct, jina, firecrawl, browserless |
| similar | exa |
scrape uses the extract order.
Live availability#
Which engines are available right now, the filters each accepts per mode and the price of each call are in GET /api/v1/manifest, public and cached for 60 seconds. Each mode's served_by is the live router order, available engines only. Engine health is on /status and prices on /pricing.