⚖️ AgentCourt

Policy-driven dispute resolution for AI agent commerce. When agents transact and things go wrong — deterministic rulings in under 500ms.

Try the API → GitHub Quick Start

Policy Templates

Deterministic Rules

<500ms

Ruling Speed

$0.05

Per Dispute

Integration Paths

Available Policies

api-quality freelance-delivery milestone-payment bug-bounty sla-monitoring scope-dispute physical-commerce

File a Dispute in 30 Seconds

curl -X POST https://agentcourt-api-production.up.railway.app/v1/disputes \
  -H "Content-Type: application/json" \
  -d '{
    "policy": "api-quality",
    "claim": "API returned XML instead of JSON",
    "desired_remedy": "full_refund",
    "metadata": {"response_received": true, "schema_matches": false}
  }'

Integration Paths

🐍 Python SDK

Zero dependencies, stdlib only. pip install agentcourt

📜 JavaScript SDK

Native fetch, TypeScript definitions. npm install @agentcourt/sdk

🔌 MCP Server

6 tools for Claude & Cursor. File disputes from your AI assistant.

🤖 ElizaOS Plugin

FILE_DISPUTE action for ElizaOS agents. Native runtime integration.

🌐 REST API

OpenAPI 3.1 spec, Postman collection, curl-friendly. No SDK needed.

⛓️ x402 Native

$0.05/dispute in USDC on Base. Free tier: 100/month.

Why Deterministic?

Same input → same output. Every time. That's the foundation for trust scoring, precedent, and audit trails in agent commerce.

No Hallucination Risk

Rules evaluate booleans, numbers, dates. No LLM in the ruling path means no hallucinated verdicts.

Sub-500ms Latency

Fast enough to block a bad transaction before it completes. LLM evaluation takes 5-30 seconds.

Full Audit Trail

Every ruling includes the matched rule ID, confidence score, and reasoning. Testable and replayable.