agent-verify (atool-proof)

Fact verification and claim attestation for AI agents.

Ask a question. Get a verdict, confidence score, and sources.

What it does

agent-verify lets your agent verify factual claims against live web evidence, with structured verdicts and confidence scores.

curl -X POST https://api.agenttool.dev/v1/verify \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"claim": "The Sun is approximately 93 million miles from Earth"}'

{
  "verdict": "disputed",
  "confidence": 0.5,
  "sources": [...6 sources...],
  "caveats": ["The distance is an average — Earth orbit is elliptical"]
}

Verdicts

Verdict	Meaning
verified	Claim supported by multiple independent sources
disputed	Claim is partially true or context-dependent
false	Claim contradicted by evidence
unverifiable	Insufficient evidence to assess

Endpoints

Method	Path	Description
POST	/v1/verify	Verify a single claim
POST	/v1/verify/batch	Verify multiple claims
GET	/health	Health check

SDK

pip install agenttool-sdk

from agenttool import AgentTool

at = AgentTool()
result = at.verify.claim("The Eiffel Tower is in Paris")
print(result.verdict, result.confidence)  # verified  0.99

Tech stack

• Hono + TypeScript + Bun
• OpenAI GPT-4o for evidence assessment
• Google Search (via SerpAPI) for live evidence gathering
• Deployed on Fly.io (London)

Get started

1. Create a free project at app.agenttool.dev
2. Free tier: 5 verifications/day. Paid from $29/mo.

---

Part of agenttool.dev — memory, tools, verify, economy, traces. One API key.