Search

Call the Talarion search endpoint directly over HTTPS — retrieve verified assertions with a single POST.

Your first request

bash

curl -X POST https://api.talarion.com/v1/search \
  -H "Authorization: Bearer $TALARION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "What are the most significant AI model releases in 2026?"}'

python

import os
import requests
 
resp = requests.post(
    "https://api.talarion.com/v1/search",
    headers={"Authorization": f"Bearer {os.environ['TALARION_API_KEY']}"},
    json={"query": "What are the most significant AI model releases in 2026?"},
)
resp.raise_for_status()
 
for fact in resp.json()["results"]:
    print(f"{fact['question']} → {fact['answer']} (true by {fact['true_by']})")
    for url in fact["sources"]:
        print(f"    source: {url}")

typescript

const res = await fetch("https://api.talarion.com/v1/search", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.TALARION_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    query: "What are the most significant AI model releases in 2026?",
  }),
});
 
const { results } = await res.json();
console.log(results);

Response

A JSON object with a results array. Each entry is a self-contained, current fact — correcting a misconception the querying LLM would otherwise be operating under — with a question, an answer, true_by (the date the fact was known true — ISO YYYY-MM-DD, or null when the source carries no truth-time), and sources, a list of the upstream URL(s) the fact is drawn from. By default entries are ordered with the most recent true_by first; set timeline: false to order by relevance instead. A response is never empty; a question returns the nearest available facts. See Overview for how to use them.

json

{
  "results": [
    {
      "question": "What flagship AI model did Anthropic release in May 2026?",
      "answer": "Claude Opus 4.8.",
      "true_by": "2026-05-15",
      "sources": ["https://www.anthropic.com/news/claude-opus-4-8"]
    },
    {
      "question": "What is the estimated total amount major tech companies will spend on AI data centers in 2026?",
      "answer": "$650 billion.",
      "true_by": "2026-03-01",
      "sources": [
        "https://www.cnbc.com/2026/03/01/big-tech-ai-data-center-spending-2026.html",
        "https://www.reuters.com/technology/ai-capex-2026-forecast/"
      ]
    }
  ]
}

The shape is built for an LLM to consume rather than a person to read, with terse question/answer pairs stamped with a date and sources grounding a model most reliably.

Parameters

Field	Type	Required	Description
`query`	string	Yes	The search question you want caught up on — e.g. “How is insurance coverage of GLP-1 obesity drugs changing?”. One focused question about one topic retrieves best.
`model_id`	string	No	Your model as an OpenRouter id (e.g. `anthropic/claude-opus-4.8`). Filters results to facts that postdate your model’s training cutoff.
`k`	integer	No	Number of facts to return. Defaults to `20`; clamped to the range `10`–`100`.
`timeline`	boolean	No	Result ordering. `true` (default): most recent `true_by` first. `false`: ordered by relevance.
`true_by_min`	datetime	No	Only return facts with `true_by` on or after this ISO 8601 datetime. Overrides the `model_id`-derived training-cutoff floor. Naive datetimes are treated as UTC.
`true_by_max`	datetime	No	Only return facts with `true_by` on or before this ISO 8601 datetime — e.g. for date-boxed backtesting. Must be after `true_by_min`.

Errors & limits

Status	Meaning
`401`	Missing or invalid API key.
`402`	Free-tier monthly call limit reached (1,000 calls/month). Upgrade in Billing.
`422`	Invalid request body — e.g. an empty `query`, or `true_by_max` not after `true_by_min`.