Api x402 offerings

unichain address snapshot (0x..40hex): balance_wei, nonce, is_contract on a chainId-verified RPC.

Fetch 1-5 URLs to clean markdown/text (64KB cap/page, SSRF-guarded, sha256 each). $0.005.

Probe one URL ONCE right now with the shared, SSRF-guarded x402 probe engine and compare the observation against that URL's ACTIVE SLA declaration (expected_p95_ms overrides the declared threshold for a single call). Verdict ok | breach{declared, observed}. Single-probe honesty: the response and any certificate state the observation basis verbatim ('single probe at <ts>') with sample_size 1, never a statistical claim; uptime_target is recorded in the declaration but is not evaluated from one probe. A breach WITH a payment_ref (<= 100 chars) issues an HMAC-SHA256 signed BreachCertificate carrying {sla_declared, observed{reachable, latency_ms}, probe_evidence{envelope_sha256, probe_ts}, payment_ref} so the parties can settle the refund out-of-band under their own custody. DDG holds no funds, executes no refund, moves nothing: certificates are evidence only.

ERC-20 balanceOf + metadata (symbol/decimals/name) on any supported chain, gate-verified eth_call, exact formatted amount.

Check a candidate transaction against an agent spend policy; signed allow|deny|needs_approval verdict with fired rule and headroom.

optimism tx lookup (0x..64hex): eth_getTransactionByHash + receipt; tx_not_found is a valid $0.01 verdict.

Get SSL certificate details for a hostname: issuer, expiry, SANs, fingerprint.

One gate-verified keyless read op on one chain (ethereum, base, polygon, arbitrum, optimism, avalanche, bnb, unichain). Read-only method allowlist, chainId-verified endpoints, local reth first for Ethereum when healthy. 1 MiB cap, 8s timeout.

Encrypt a payload to any published agent PGP key. Only their private key can decrypt.

Live API-contract drift probe: fingerprint the live OpenAPI doc vs a pinned fingerprint. match|drifted|unreachable.

worldchain read-only RPC: allowlisted no-param reads or bounded caller method (eth_blockNumber/eth_chainId/... ). 1MiB cap.

Check whether an agent-held receivable is factoring-ready and get a signed ready|not_ready verdict with blocking findings.

Translate text between languages. Self-hosted LibreTranslate or cloud fallback.

Verify an A2A agent card before trusting or routing to it: dual-probe /.well-known/agent-card.json and agent.json, check MUST fields (name/description/url/version/capabilities), classify protocolVersion, flag the known 0.3<->1.0 silent schema breaks, verify any JWS signature over RFC 8785 JCS canonical bytes (did:web key resolution), and live-probe the advertised interface with tasks/send. Per-step verdict + remediation.

ERC-20 metadata on any supported chain: name/symbol/decimals/totalSupply via gate-verified eth_call.

unichain tx lookup (0x..64hex): eth_getTransactionByHash + receipt; tx_not_found is a valid $0.01 verdict.

Proxy a third-party API call through DDG: payload redacted en route, auth headers stripped, response returned with a server-signed attestation binding original-vs-sent sha256. Your agent's prompts never leak.

Thread view between two agents - negotiation history across the encrypted channel.

Return the best cross-venue prediction-market spread.

Audit an MCP server or tool surface for installability, schema quality, auth, and prompt-injection risk. Fully automated: delivered inline in the paid call (no operator queue).

Check that effective authority = robot-identity AND user-delegation (intersection, not union), and that untrusted inputs (detected QR codes, voice commands) cannot authorize physical action. Maps to the confused-deputy problem.

Zero-key enrichment from free public APIs (DNS, SSL cert, CT subdomains, email validation, Gravatar).

Sanctions screen with PINNED list versions (OFAC SDN consolidated CSV + EU consolidated list): verdict clear|match + per-list version/date/rows/matches; wallet remarks scan included. Ed25519-signed receipt.

Detect double-actuation risk in a robot command pipeline: missing idempotency keys, no dedup store, and non-reversible physical actions without single-execution enforcement. One network retry should never equal two real-world effects.