What we offer
Paid x402 endpoints for agents: reliability scores, curated data feeds, JSON-schema validation, AI extraction, attested redaction, external-service preflight checks, and prepaid wallet-scoped state storage. Endpoints that accept your content enforce this reject-first policy — we are not a data broker and do not resell personal information.
How we handle sensitive data
We reject PII at the edge. Before any x402 charge settles, every
payload is scanned with format validation (Luhn, key formats, token shapes), key and
token shape matching, and Shannon-entropy heuristics. Detected payment cards, government
IDs, private keys, API secrets, JWTs, embedded files, and opaque high-entropy blobs are
blocked — the call fails with 422 screening_rejected, you are not billed,
and the content is not stored or forwarded.
Final guard: State Hub documents that pass the edge screen may receive an additional asynchronous review by a third-party AI governance model. That review exists to quarantine anything the edge missed — flagged documents return HTTP 451 and are removed from service. It is not used to collect, profile, or sell data.
Extract: only edge-cleared text is sent to a third-party model to perform the extraction you paid for. Screening runs first; matched sensitive content never leaves our platform.
Edge detectors we enforce
- Private keys, API tokens, JWTs, and payment card numbers
- Social Security numbers and similar regulated identifiers
- Embedded files and opaque high-entropy secrets
- Email addresses and phone numbers (rejected on Extract; controlled on State Hub)
On State Hub, hard secrets are rejected at the edge; email and phone are allowed only after edge clearance.
What we do
We provide scrubbed, wallet-isolated working memory for autonomous agents — not a personal-data warehouse. After edge screening clears a payload, State Hub stores JSON the agent needs to resume work across sessions: task progress, tool outputs, structured facts, and other operational state that has been checked for secrets and regulated identifiers before it ever lands in our database.
- Business case: long-running agents need durable memory without holding signing keys or shipping raw PII on every x402 call. Prepaid State Hub credits let agents save, query, and delete scrubbed state at micro-op prices instead of paying per-settlement overhead for every read/write.
- Safety by design: only edge-cleared documents are stored; hard secrets never enter; async AI governance quarantines anything that slips through. Data is TTL-bound, quota-capped, and purgeable — working memory, not permanent records.
- Isolation: each paying wallet sees only its own keys and documents (database-enforced row-level security). We do not merge, enrich, or sell agent state across customers.
Stateless endpoints (scores, feeds, validate, redact) never retain your payload beyond the request. State Hub is the one surface where scrubbed, anonymized operational data persists — because that is what agents pay us to provide safely.
What we do not do
- We do not collect personal data to resell or monetize it.
- We do not store or forward payloads that fail edge screening.
- We do not serve quarantined State Hub documents (HTTP 451).
Before you pay
Use this page or GET /v1/policy to review
size limits, detector definitions, and rejection codes. A failed screen returns
422 screening_rejected and does not settle x402 payment.
Accepted shapes
- JSON values: objects, arrays, strings, numbers, booleans, null.
- Human-readable text and structured records under the per-doc size limit.
Endpoint rules (detail)
| Endpoint | Where data goes | Screening behavior |
|---|---|---|
POST /v1/state/save |
Stored after edge clearance | Edge screen — opaque blobs: rejected at the edge before billing. Edge screen — secrets / regulated PII: rejected at the edge before billing. Edge screen — email / phone: allowed only after edge clearance. Final guard: async AI governance review by anthropic — quarantine flagged docs; we do not resell PII. |
POST /v1/extract |
Edge-cleared text to third-party model | Edge screen — opaque blobs: rejected at the edge before billing. Edge screen — secrets / regulated PII: rejected at the edge before billing. Edge screen — email / phone: rejected at the edge before billing. After edge clearance only: content is sent to anthropic to perform the paid extraction — not for data collection. |
Detectors
| Id | Label | Class | Description |
|---|---|---|---|
base64_media |
Embedded media blob | opaque | Base64-encoded image/video (PNG, JPEG, WebP, MP4). Store a URL or hash instead. |
high_entropy |
High-entropy / opaque blob | opaque | A long string whose Shannon entropy exceeds the opacity threshold (encrypted/compressed/packed data we cannot screen). |
private_key |
Private key material | secret_or_pii | A 32-byte hex string resembling an EVM/crypto private key. |
jwt |
JSON Web Token | secret_or_pii | A bearer JWT (header.payload.signature). |
aws_key |
AWS access key id | secret_or_pii | An AKIA-prefixed AWS access key identifier. |
credit_card |
Payment card number | secret_or_pii | A 13–19 digit, Luhn-valid card number. |
ssn |
US Social Security Number | secret_or_pii | A US SSN in NNN-NN-NNNN form. |
email |
Email address | soft_identifier | An RFC-style email address. |
phone |
Phone number | soft_identifier | A North-American-format phone number. |
Limits
- State doc max: 65,536 bytes
- Extract content max: 100,000 characters
Rejection codes
| Status | Error | Meaning |
|---|---|---|
| 413 | payload_too_large | Doc/content exceeds the size limit. |
| 413 | quota_exceeded | Per-wallet row or byte quota would be exceeded. |
| 422 | validation_error | Body failed schema validation (bad key, missing fields, wrong types). |
| 422 | screening_rejected | Content tripped a reject-class detector (see screening.detectors). |
| 422 | cost_guard_exceeded | Projected model token cost exceeds the price margin floor (extract). |
| 451 | doc_quarantined | Document was quarantined by async review and is no longer served. |