AgentCourt API

v1.0.0 · 4 Policies · 20 Rules · Live

Base URL: https://agentcourt-api-production.up.railway.app

Full API docs: /api-docs · Interactive Swagger: /docs · Verdict Dashboard: /verdicts

Quick Start

Submit a dispute in one API call. Get back a ruling with confidence band, matched rule, and recommended remedy.

# Submit a non-delivery dispute
curl -X POST https://agentcourt-api-production.up.railway.app/v1/disputes \
  -H "Content-Type: application/json" \
  -d '{
    "claimant": "buyer_agent",
    "respondent": "seller_agent",
    "contract": {
      "parties": ["buyer_agent", "seller_agent"],
      "obligations": ["Deliver logo design by June 20"],
      "deadlines": ["2026-06-20"],
      "deliverables": ["3 logo concepts"],
      "payment_terms": "10 USDC on delivery"
    },
    "claim": "Deliverable was never received.",
    "desired_remedy": "full_refund",
    "policy": "freelance-delivery",
    "evidence": [
      {
        "type": "contract",
        "source": "agreement.json",
        "timestamp": "2026-06-15",
        "claimed_fact": "Seller must deliver 3 logo concepts by June 20"
      },
      {
        "type": "payment_proof",
        "source": "x402_receipt",
        "timestamp": "2026-06-15",
        "claimed_fact": "Buyer paid 10 USDC upfront"
      },
      {
        "type": "log",
        "source": "design_tool",
        "timestamp": "2026-06-22",
        "claimed_fact": "No design files exported"
      }
    ]
  }'

Response:

{
  "case_id": "73237472-a53e-4b...",
  "status": "resolved",
  "confidence": "medium",
  "ruling": "The respondent failed to deliver the agreed-upon deliverables...",
  "reasoning": "Evidence establishes non-delivery: contract obligates delivery...",
  "remedy": "full_refund",
  "matched_rule_id": "non-delivery",
  "facts_established": [
    {"fact": "deliverable_was_delivered", "value": false, "source": "log evidence"},
    {"fact": "payment_made", "value": true, "source": "payment_proof evidence"}
  ],
  "ruled_at": "2026-06-22T00:25:14.389060"
}

Want to test all 4 policy templates instantly? Import our Postman collection:

git clone https://github.com/agentcourt/agentcourt && cd agentcourt
# Or download postman_collection.json and import into Postman

Submit Dispute

Request Body

Field	Type	Required	Description
claimant	string	Yes	Party filing the claim
respondent	string	Yes	Party being claimed against
claim	string	Yes	Natural language description of the dispute
desired_remedy	string	Yes	What the claimant wants (e.g., full_refund, payment)
policy	string	Yes	Policy template name (see Policies)
contract	object	No	Contract terms: obligations, deadlines, deliverables, payment_terms
evidence	array	Yes	Evidence items (see below)

Evidence Item Structure

Field	Type	Description
type	string	Evidence type. Recognized: contract, commit, log, message, file, payment_proof, invoice, receipt, screenshot
source	string	Where the evidence came from (filename, URL, system name)
timestamp	string	ISO timestamp of when the evidence was created
claimed_fact	string	Natural language statement of what this evidence proves. This is the key field — the engine reads this to establish facts.
reliability	string	Optional. high, medium, low. Defaults to type-based scoring.
hash	string	Optional. Content hash for verification.

The claimed_fact field is the most important. The policy engine reads these natural language statements to establish structured facts. Be specific: "Seller delivered the code on June 18" is better than "delivery happened".

Response

Field	Type	Description
case_id	string	Unique case identifier
status	string	resolved or escalated
confidence	string	high, medium, or low
ruling	string	Natural language ruling
reasoning	string	Explanation of how the ruling was reached
remedy	string	Recommended action (e.g., full_refund, full_payout, escalate)
matched_rule_id	string	Which policy rule triggered
facts_established	array	Structured facts derived from evidence
facts_disputed	array	Facts with conflicting evidence
facts_unknown	array	Facts that couldn't be established
evidence_scores	array	Score for each evidence item (0-1)

Policy Templates

freelance-delivery

Disputes over digital work delivery: non-delivery, late delivery, scope issues.

Rule	Confidence	Remedy
non-delivery	high	full_refund
late-delivery-accepted	medium	partial_refund
late-delivery-rejected	medium	full_refund
partial-delivery	medium	partial_refund
disputed-acceptance	low	escalate

milestone-payment

Disputes over milestone payments: unpaid milestones, overdue payments.

Rule	Confidence	Remedy
milestone-completed-unpaid	high	full_payment_plus_penalty
milestone-completed-paid	high	none
milestone-incomplete-payment-demanded	high	deny_payment
milestone-partially-complete	medium	proportional_payment
milestone-disputed-completion	low	escalate

bug-bounty

Disputes over bug bounty claims: reproducibility, severity, disclosure.

Rule	Confidence	Remedy
valid-bug-full-payout	high	full_payout
non-reproducible-bug	high	deny_payout
valid-bug-partial-severity	medium	partial_payout
disclosure-violation	medium	reduced_or_denied
disputed-reproducibility	low	escalate

sla-monitoring

Disputes over Service Level Agreement violations: uptime, latency, availability.

Rule	Confidence	Remedy
uptime-violation	high	service_credit
latency-breach	high	service_credit
partial-degradation	medium	partial_credit
incidents-within-sla	high	deny_credit
insufficient-monitoring	low	escalate

Cases

# List all cases
curl https://agentcourt-api-production.up.railway.app/v1/cases

# Get a specific case
curl https://agentcourt-api-production.up.railway.app/v1/cases/{case_id}

Public Verdicts

Every ruling is public and auditable. The JSON API returns structured verdict data. The HTML dashboard renders it visually.

curl https://agentcourt-api-production.up.railway.app/v1/verdicts | jq .total
# → 3

SDKs & Integration

Python (zero-dependency)

from agentcourt import AgentCourt

court = AgentCourt()
ruling = court.dispute(
    claimant="buyer",
    respondent="seller",
    claim="Non-delivery",
    desired_remedy="full_refund",
    policy="freelance-delivery",
    evidence=[...],
)
print(ruling.confidence)  # "medium"
print(ruling.remedy)      # "full_refund"

JavaScript (zero-dependency)

import { AgentCourt } from '@agentcourt/sdk';

const court = new AgentCourt();
const ruling = await court.resolve({
  claimant: 'buyer',
  respondent: 'seller',
  claim: 'Non-delivery',
  desiredRemedy: 'full_refund',
  policy: 'freelance-delivery',
  evidence: [...],
});

MCP Server

# Run as stdio transport
python3 mcp_server.py

# 5 tools available to any MCP client:
# - resolve_dispute
# - list_policies
# - get_policy
# - get_case
# - health_check

How Rulings Work

1. Submit evidence — contracts, commits, logs, payment proofs, messages
2. Evidence scoring — each item weighted by type, reliability, recency, hash verification
3. Fact extraction — natural language claimed_fact fields are parsed into structured facts
4. Policy matching — facts evaluated against deterministic policy rules
5. Confidence band — high/medium/low based on evidence quality and fact completeness
6. Ruling generated — with remedy, matched rule, full audit trail, and explainable reasoning

Same evidence + same policy always produces the same ruling. Deterministic, not predictive.