Call content
90 days
The inputs your agent sent, the answer, any error message, the audit trail and the source details. Cleared after 90 days.
Security and data
What happens to the data your agent sends
Where calls run, what Strale keeps and for how long, where the data comes from, and how each tool is checked.
- Calls processed in
- United States
- Call content cleared after
- 90 days
- Call records kept for
- 1,095 days
- API keys stored as
- SHA-256 hashes
Where calls are processed
Strale’s API and database run in the United States. Every call is processed there, and each call’s audit record names where it ran.
A tool that answers from an outside source sends that source the inputs it needs for the answer. The vendors are listed below, and each tool’s page names the source of its data.
What is kept, and for how long
Every call leaves a record, and the record has two parts on two clocks.
The rest of the record
1,095 days
Which tool ran, when, the outcome, the price, how long it took, and the integrity hash that links it to the record before it. Kept for 1,095 days (3 years).
Content is cleared in place, so the chain of integrity hashes stays verifiable after the content is gone. A dispute raised within 90 days can be checked against what the call actually sent and returned; after that, the record still shows that the call ran, when, and with which tool.
A record under legal hold is kept until the hold ends.
Keys and payments
API keys are stored only as SHA-256 hashes. The full key is shown once, when it is created. Strale cannot show it again: a lost key is replaced, never recovered.
Cards are handled by Stripe. Balance top-ups run through Stripe Checkout, so card details go to Stripe and never reach Strale.
x402 payments are settled by the Coinbase x402 facilitator (USDC on Base). Your agent’s wallet signs each payment itself; Strale never holds its keys.
Sub-processors and data vendors
The services that handle screening, payments, page rendering and logs for Strale, and the vendors behind some company-data tools. Your agent has no account with any of them: Strale holds the relationships.
| Used for | Vendor |
|---|---|
| Sanctions screening | Dilisense |
| Politically exposed person screening | Dilisense |
| Adverse-media screening | Dilisense |
| Adverse-media screening, fallback search | Serper.dev (Google) |
| Tool search | Anthropic Claude (search ranking; no embeddings) |
| AI risk summaries | Anthropic Claude |
| Rendering web pages | Browserless.io |
| Card payments | Stripe |
| x402 payments | Coinbase x402 facilitator (USDC on Base) |
| Operational logs | Better Stack |
| US company records | Cobalt Intelligence |
| US employer ID checks | Liberty Data |
| Company ownership links | GLEIF L2 |
| French court and insolvency notices | BODACC |
The same list is served by the API at /v1/platform/facts. Each tool’s page names the source of its own data.
Where the data comes from
Strale goes to the source of record first: official registers and APIs, then licensed data from vendors whose terms allow resale, then direct reads of public pages, one request at a time. The rules are the same for every tool.
No logins, no paywalls, no fake or borrowed accounts. Strale reads what is public, logged out.
A block is an answer. A refusal, a challenge page or a robots.txt rule ends the request. Strale never solves CAPTCHAs, rotates addresses or retries a blocked request another way.
Every request says who it is. Strale identifies itself by name in each request it sends to a website.
No harvesting people data. Strale does not collect profiles or contact details of people. Personal data comes back only where a court or registry publishes it, as published, for the one request that asked. Sanctions, PEP and adverse-media screening runs on licensed data, for compliance checks.
No sites that forbid automated access. Platforms that forbid it and enforce it, and commercial data aggregators, are on a blocklist Strale checks before every fetch.
Pages your agent names are read once. When your agent asks Strale to read a page, Strale fetches that one public page for that one request, under the same rules.
For each answer, Strale records which source the data came from and how it was obtained.
How tools are checked
Strale runs its own test suites against its tools. Which suites a tool has depends on what it does:
- Known answer
- Calls the tool with an input whose answer is known, and checks the fields that must always be there.
- Schema
- Checks the answer has the shape the tool promises.
- Negative
- Sends bad input and checks the tool refuses it cleanly.
- Edge case
- Sends unusual but valid input and checks the tool still answers correctly.
- Dependency health
- Checks that the tool’s upstream source is reachable.
Some suites call the real source; others replay a saved baseline answer to check Strale’s own handling of it. Tools whose source costs nothing to call are tested on a schedule. Tools that pay a vendor per call are watched through their real traffic: the outcome of customers’ calls feeds their health record.
Each tool’s page shows the date of its last recorded test and the share of its tests that passed on that latest run. A tool with no recorded test says so. A test result tells you what the tool returned on that date; it is not a score.
When a tool’s real calls fail too often, Strale takes it out of the catalogue, and it comes back once a fix passes its checks.
Browse the toolsDisputing a result
You can ask for a person to review the result of any call made with an API key. Send the reason, and optionally the field you dispute, to the call’s dispute endpoint. A person at Strale reviews every dispute.
POST /v1/transactions/:id/dispute
curl -X POST https://api.strale.io/v1/transactions/TRANSACTION_ID/dispute \
-H "Authorization: Bearer $STRALE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"reason": "The register shows this company as dissolved.", "affected_field": "status"}'The account holder authenticates with the API key. Someone who was given the call’s shareable audit link can dispute with the link’s token instead, and adds a contact_email so Strale can reply.
GDPR Article 22 classification
Every call’s audit record states how the tool’s output relates to automated decisions about people, and gives the dispute endpoint. The customer making the call is the controller for any personal data in it; Strale processes it on the customer’s behalf.
data_lookup- Returns facts from a source, and is not a decision. 292 tools have this class.
screening_signal- Reports matches against lists or databases, such as sanctions lists. The match is a fact about the source; acting on it automatically, without review, is where Article 22 applies. Tools with this class: Adverse Media Check, Insolvency Check, PEP Check, Sanctions Check, UK Disqualified Director Check.
risk_synthesis- An AI model writes a risk summary from other checks’ results. A workflow that acts on it automatically is an automated decision under Article 22. Tools with this class: Risk Summary Writer.
A bundled check takes the highest class among its steps.
Contact
Security reports, data requests and questions about anything on this page go to one address, read by a person.
Email [email protected]