Low-value, failure-tolerant
Crawl access, API calls, MCP tools. Standardised, cheap, easy to retry.
The credit layer for the agentic economy
x402 proves agents got paid. Nothing proves they were worth paying. Assay measures what every paid endpoint actually delivers — then bonds it, so agents get recourse when it doesn't.
The gap
Agents pay first and hope. If an endpoint takes the money and returns nothing, times out, or returns garbage, that's the agent's loss. Fine at $0.003. Useless at $5.
Crawl access, API calls, MCP tools. Standardised, cheap, easy to retry.
Inference routing, agent-to-agent delegation, data purchases, commerce. All need recourse. None have it.
How Assay works
Probers pay endpoints and record what came back. The results are published free as the reliability index for agent services.
Endpoints stake to earn guaranteed-delivery status. Slashed on non-delivery, off-spec response, or overcharge — every trigger mechanically decidable.
Capital supplies the bonds and earns the premium. Yield in USDC, from fees paid for a real service — single-digit to low-teens APY, net of losses.
The reliability index
The reliability index for x402 services. We pay endpoints, record what came back, and publish it — free. Bonded endpoints refund your agent when they don't deliver.
| Endpoint | Category | Delivered · 30d | p99 | Price | Failure prob. | Status |
|---|---|---|---|---|---|---|
| infer.routerhub.ai/v1/chat | Inference | 820ms | $0.42 | 0.6% | Bonded | |
| data.tickstream.io/quotes | Data | 140ms | $1.20 | 0.9% | Bonded | |
| mcp.ledgerlens.xyz/tools | MCP tool | 310ms | $0.03 | 2.2% | Probing | |
| crawl.pagevault.io/fetch | Crawl | 1.2s | $0.004 | 3.8% | Probing | |
| gen.pixelforge.run/image | Inference | 4.1s | $0.28 | 7.1% | Watch | |
| agent.dealdesk.app/quote | Delegation | 6.9s | $5.00 | 12.4% | Watch |
No endpoints match your search. Try another name or status.
How we price risk
01
Beta-Binomial posterior with a pessimistic prior. Nobody starts at 0%.
02
Rising p99 latency flags timeouts before they happen; value-split failure rates catch selective defection.
03
300 endpoints in one cloud region is one risk, not 300. Hard caps enforced.
04
Anchored to peak daily revenue, so nobody builds history small and defects big.
Roadmap
Months 0–6 · Now
Probes run in-house, published free.
Months 6–12
Paid operators, settled in USDC.
Month 12+
Staked probers, auto-slashed.
Stake a bond, get the badge agents route to. Premium ≈ expected loss × (1 + load), paid in USDC.
Bond an endpointSupply bond capital, earn the premium. Single-digit to low-teens APY, net of losses — unglamorous, and still there in two years.
Join the waitlistJoin the agentic economy
Get the index as it grows — new endpoints, loss curves, and the first bonded services.