⚡ 200ms average response time

Disputes, resolved in 200 milliseconds.

Policy-driven dispute resolution for AI agent commerce. Submit evidence. Get a binding ruling. Deterministic, transparent, built for autonomous transactions.

No signup required · curl and go · MIT licensed
POST /v1/disputes
curl -X POST https://api.agentcourt.to/v1/disputes \
  -H "Content-Type: application/json" \
  -d '{
    "policy": "freelance-delivery",
    "claim": "Non-delivery of website",
    "evidence": [{
      "source": "github",
      "fact": "No commits in 30 days"
    }]
  }'

# 200 OK — 187ms
{
  "status": "ruled",
  "ruling": "Respondent failed to deliver.
   Full remedy owed to claimant.",
  "matched_rule": "non_delivery",
  "confidence": 0.92
}
200ms
avg ruling time
34
active rules
$0.50
per ruling (USDC)
100%
deterministic
How it works

Three steps to a binding ruling.

No LLM. No prompts. Just deterministic policy evaluation.

1

Submit your dispute

POST your contract, claim, and evidence. Specify which policy applies.

POST /v1/disputes
policy: "freelance-delivery"
2

Engine evaluates

Evidence is scored 0.0–1.0. Facts are extracted. Rules are matched deterministically.

evidence_score: 0.85
matched_rule: non_delivery
3

Ruling delivered

Structured ruling with remedy, confidence score, and full audit trail. Stored for precedent.

status: "ruled"
confidence: 0.92
Policy templates

Battle-tested dispute policies.

Use them as-is or customize. Every policy is transparent and auditable.

Freelance Delivery

Work-for-hire contracts with deadlines, deliverables, and quality assessment.

6 rules

Milestone Payment

Staged deliverables with payment gates and completion tracking.

5 rules

Bug Bounty

Severity disputes, reproducibility requirements, disclosure compliance.

5 rules

SLA Monitoring

Uptime, latency, and availability violations with automatic credit calculation.

5 rules

API Quality

Wrong data, schema mismatches, and service errors on paid API calls.

7 rules

Physical Commerce

Product disputes: damage, non-delivery, wrong items, and returns.

6 rules
Developer SDKs

Integrate in five minutes.

Python, JavaScript, curl, or MCP server. Pick your weapon.

Python
JavaScript
curl
resolve_dispute.py
from agentcourt import AgentCourt

court = AgentCourt()  # uses api.agentcourt.to

ruling = court.resolve_dispute(
    policy="freelance-delivery",
    claim="Non-delivery of website",
    desired_remedy="Full refund of $5,000",
    evidence=[{
        "source": "github",
        "fact": "No commits for 30 days"
    }]
)

print(ruling.status)   # → "ruled"
print(ruling.ruling)   # → "Respondent failed..."
print(ruling.confidence) # → 0.92
FAQ

Questions, answered.

What is AgentCourt?

A policy-driven dispute resolution API for AI agent commerce. Submit evidence, get a deterministic ruling in 200ms. No LLM, no hallucination.

How does it work?

POST your contract, claim, and evidence. The engine scores evidence (0.0–1.0), extracts facts, matches policy rules deterministically, and returns a structured ruling.

How much does it cost?

$0.50 USDC per ruling via x402 payment protocol on Base network. Self-hosting is free under the MIT license.

Is it deterministic?

Yes. The same evidence always produces the same ruling. The policy engine uses rule-based evaluation, not generative AI. Every ruling cites the specific rule matched.

Can AI agents use it directly?

Yes. AgentCourt exposes an MCP server, OpenAI function calling schema, and a /.well-known/agent.json endpoint for agent discovery. Any agent can submit and receive rulings autonomously.

What happens to my data?

All cases are persisted to a durable volume at /data. Cases survive restarts and can be retrieved by case ID for appeals and precedent.

Start resolving disputes now.

Read the docs, grab an SDK, submit your first dispute. No signup, no API key, no credit card.

View API Reference →