Fact-check any claim with AI-powered evidence gathering, multi-source verification, and confidence scoring. Get a verdict — verified, disputed, false, or unverifiable — in under 30 seconds.
Live — agent-verify is available now. Create a project to get your API key.
Base URL: https://api.agenttool.dev · All endpoints require Authorization: Bearer YOUR_API_KEY
How it works: Claims are parsed into search queries → evidence gathered from web, Wikipedia, and authoritative sources → LLM judge evaluates evidence → confidence scored with heuristics → verdict returned with sources.
Verify a single factual claim. The pipeline gathers multi-source evidence, runs an LLM judge, and returns a verdict with confidence score and cited sources.
🔒 Bearer token required| Field | Type | Required | Description |
|---|---|---|---|
| claim | string | required | The factual claim to verify. Max 2,000 characters. e.g. "The Eiffel Tower is 330 metres tall." |
| context | string | optional | Additional context that helps interpret the claim (e.g. conversation history, agent task description). Max 2,000 characters. |
| domain | string | optional | Claim domain — improves evidence sourcing. One of: finance, legal, medical, science, general. Auto-detected if omitted. |
curl -X POST https://api.agenttool.dev/v1/verify \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"claim": "The Eiffel Tower is 330 metres tall.",
"domain": "general"
}'
const result = await at.verify.check( "The Eiffel Tower is 330 metres tall.", { domain: "general" } ) // result.verdict → "verified" | "disputed" | "false" | "unverifiable" // result.confidence → 0.0 – 1.0
{
"claim": "The Eiffel Tower is 330 metres tall.",
"verdict": "verified",
"confidence": 0.91,
"evidence": {
"supporting": [
{
"source": "wikipedia",
"url": "https://en.wikipedia.org/wiki/Eiffel_Tower",
"title": "Eiffel Tower — Wikipedia",
"snippet": "The Eiffel Tower is 330 m (1,083 ft) tall...",
"reliability": 0.85,
"position": "supports"
}
],
"contradicting": [],
"neutral": []
},
"sources": [
{
"url": "https://en.wikipedia.org/wiki/Eiffel_Tower",
"title": "Eiffel Tower — Wikipedia",
"reliability": 0.85
}
],
"caveats": ["Height includes the broadcast antenna added in 1957."],
"processing_ms": 4820
}
| Field | Type | Description |
|---|---|---|
| verdict | string | See Verdicts below. |
| confidence | number | Confidence score 0.0–1.0. Above 0.75 is high confidence. |
| evidence.supporting | array | Sources that support the claim, each with url, snippet, reliability. |
| evidence.contradicting | array | Sources that contradict the claim. |
| caveats | array | Nuances, edge cases, or time-sensitivity notes. |
| processing_ms | number | Total pipeline duration in milliseconds. |
Verify up to 10 claims in a single request. Claims are processed in parallel. Useful for fact-checking agent-generated reports, validating research summaries, or screening multiple assertions at once.
🔒 Bearer token required| Field | Type | Required | Description |
|---|---|---|---|
| claims | array | required | Array of 1–10 claim objects. Each has claim (required), context (optional), domain (optional). |
curl -X POST https://api.agenttool.dev/v1/verify/batch \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"claims": [
{ "claim": "The Eiffel Tower is 330 metres tall.", "domain": "general" },
{ "claim": "Bitcoin was created in 2009.", "domain": "finance" },
{ "claim": "Water boils at 90°C at sea level.", "domain": "science" }
]
}'
const results = await at.verify.batch([ { claim: "The Eiffel Tower is 330 metres tall." }, { claim: "Bitcoin was created in 2009.", domain: "finance" }, { claim: "Water boils at 90°C at sea level.", domain: "science" } ]) // Returns array of VerificationResult — same shape as single verify
Returns an array of VerificationResult objects in the same order as the input claims. Each result has the same shape as the single verify endpoint.
Parallel processing: All claims in a batch are verified concurrently. Latency is bounded by the slowest claim, not the sum. A batch of 10 claims takes roughly the same time as 1.
The claims array is empty or contains more than 10 items.
| Verdict | Meaning | Typical confidence |
|---|---|---|
verified |
Multiple reliable sources confirm the claim. | 0.70 – 0.99 |
disputed |
Sources conflict — some support, some contradict. Requires human review. | 0.30 – 0.70 |
false |
Multiple reliable sources contradict the claim. | 0.01 – 0.40 |
unverifiable |
Insufficient evidence found, or claim is opinion/prediction. Confidence returned as 0. | 0.00 |
Automatically fact-check claims in agent-generated research reports before surfacing them to users. Flag disputed or false findings for human review.
Run LLM outputs through verify before using them downstream. Catch confident-sounding fabrications before they propagate through your agent pipeline.
Gate data pipelines on verified facts. Only pass claims with verdict == "verified" and confidence > 0.8 to downstream systems.
Use domain hints (finance, legal, medical, science) for better evidence sourcing. Finance claims check market data; medical claims prioritise clinical literature.
Claim is empty, exceeds 2,000 characters, or batch array is empty/too large (>10 items).
Missing or invalid API key.
Too many requests. Verification is rate-limited per plan.
Timeout behaviour: If evidence gathering exceeds 25 seconds, the pipeline returns verdict: "unverifiable" with a caveat rather than erroring. You always get a response.