POST
/v1/free-key
Auto-provision a Free-Tier API key. Body: {"email": "you@example.com"}. Idempotent by email. No credit card.
10 calls/month0 €/moIdempotent
Agent evaluation · Trust Pack
4-rubric scoring, byte-identical reproducibility, 38 test assertions, 113 production audits. Built for the buyer side: score your agent, prove it works, share the Trust Pack.
A Trust Pack is a reproducible evaluation of an AI agent's card. You submit your agent card (a JSON-LD describing capabilities, governance, runtime, residency), we run it through 4 rubrics — task success, policy compliance, robustness, efficiency — and emit a scored result with a reproducibility hash. The buyer (you, or your customer) replays the same input locally and gets the same hash. No "trust me, we ran it."
The output is the same JSON whether you run it now, in a week, or in 5 years. That's the point.
API
Auto-provision a Free-Tier API key. Body: {"email": "you@example.com"}. Idempotent by email. No credit card.
Evaluate an Agent Card → Score + Trust Pack + Audit-ID. Body: agent card JSON. Returns scored rubrics, trust_pack.json, trust_pack.md, reproducibility_hash.
Fetch a specific audit entry. Useful for replaying, verifying reproducibility, integrating with internal monitoring.
POST /v1/eval
{
"agent_card": {
"id": "agent.support-triage.v1",
"capabilities": [...],
"governance": {...},
"runtime": {...},
"residency": "EU"
}
}
→ 200 OK · 30 seconds · reproducibility_hash: "0x9f2c8a..."
4 rubrics
Scores the agent's declared capabilities against the tools, resources, and prompts it exposes. A capability_presence of 1.0 means every declared tool is reachable. A tool_coverage of 1.0 means the prompt structure actually triggers the tools.
Scores declared governance — who owns it, who reviews it, what policies apply. residency check (EU vs US), audit-trail presence, kill-switch availability.
Stress-tests the agent card for missing tools, contradictory residency, weak prompts, over-broad capabilities. Score 0-1; below 0.7 means the card has real gaps.
Scores the prompt size, tool count, and resource requirements. Lower footprint → higher efficiency score. Helps you ship agents that don't burn tokens on every call.
Pricing
500 calls/mo, all 4 rubrics, idempotent by email.
Get API key →1,000 calls/mo, audit-log retention, custom rubrics flag.
Choose Starter →5,000 calls/mo, custom rubrics (1 included), priority support.
Choose Pro →15,000 calls/mo, audit-log, webhooks, Slack support, multi-user.
Choose Team →All tiers share the same 4-rubric pipeline, byte-identical reproducibility, and 38 test assertions. Annual contracts save 20%. Enterprise pricing for 50K+ calls/mo.
Trust
38/38
Test assertions (20 pytest + 18 smoke). Independently re-run 2026-07-10.
113
Production audits in the public ledger (111 with schema v2).
4
Rubrics (task_success, policy_compliance, robustness, efficiency), 83-98 lines each.
3
Endpoints (public + auth + admin), all OpenAPI-documented, idempotent free-key.
0 LLM
In the evidence path. Pure logic, deterministic, replayable.
Get started
Free API key in 30 seconds. 500 calls/month, no credit card. Score your first agent card before the next sprint review.