Linkup for AI agents

Most search APIs hand the agent a list and leave the writing to it. Linkup hands the agent whichever of three things it asked for: a written answer with its sources, the raw ranked results, or a JSON object shaped to a schema the agent supplied. That last mode is what makes Linkup unusual in the network. An agent that needs a company's founding year, headquarters, and CEO as fields, not prose, can ask for exactly that and get back something it can use without parsing.

What Linkup does well

One call, one cited answer. The default sourcedAnswer output returns a synthesized answer with the sources it drew from, and includeInlineCitations places them in the text. For a question with an answer, this is the whole job at a cent.

Structured output. Pass outputType: "structured" with a JSON-schema string in structuredOutputSchema and the response is an object matching it. This is the difference between a search and an extraction, and it removes a parsing step the agent would otherwise do with a model call.

Raw results when you want them. outputType: "searchResults" returns ranked hits for pipelines that do their own reading.

Deep search for hard questions. linkup/deep-search runs a multi-step retrieval for questions where the answer is not on any single page. It costs five cents and it is for the questions that earn it.

Date and domain controls. fromDate and toDate, include and exclude domains, and optional images.

Actions and prices through Vaaya

Priced before the call, billed only on success, under max_cost_cents.

A real call

{
  "service": "linkup",
  "action": "search",
  "params": {
    "q": "Tempo blockchain mainnet launch date and design partners",
    "outputType": "structured",
    "structuredOutputSchema": "{\"type\":\"object\",\"properties\":{\"launch_date\":{\"type\":\"string\"},\"partners\":{\"type\":\"array\",\"items\":{\"type\":\"string\"}}},\"required\":[\"launch_date\",\"partners\"]}"
  },
  "max_cost_cents": 1
}

A JSON object with a launch date and a partner list, for a cent. A deep search with a one-cent ceiling is refused before it runs, with the five-cent price returned.

When to pick Linkup, and when not to

Pick Linkup when the deliverable is an answer or a structured record rather than a reading list. Pick deep search when a question has failed at a single pass and the answer is worth five cents.

Pick Exa when recall on a descriptive query matters most, or for sweeps of many results. Pick Tavily when relevance scores or a news and finance topic filter matter. Pick Perplexity when the recency of each page decides the answer. Pick Valyu's academic search when the answer is a paper.

Gotchas

Structured output is only as reliable as the sources the search found; validate required fields before trusting them downstream. Standard search is one-pass; if the answer needs two hops, go straight to deep search rather than retrying. Keep structuredOutputSchema small; a large schema pushes the model to fill fields it did not find.

Linkup's documentation is at linkup.so. The search recipe is at /recipes/search-scrape.

Questions

What does the Linkup API return?

Your choice per call: a synthesized answer with sources, a ranked list of results, or a structured JSON object matching a schema you pass. Inline citations and images are optional. Deep search runs a multi-step retrieval for questions a single pass cannot answer.

How much does Linkup cost through Vaaya?

One cent per standard search regardless of output type. Five cents per deep search. Priced before the call, billed only on success, under the maximum cost the agent states.

Linkup or Tavily for a written answer?

Both return an answer in one call. Linkup adds a structured output mode that returns JSON matching your schema, and a deep-search tier for multi-hop questions. Tavily adds relevance scores per result and a news or finance topic filter. Pick by which of those the task needs.