Nothing sensitive leaves without a name next to it.
Keelra moves no money and assumes no AML obligations. Regulatory responsibility stays with you.
Six things your lawyer and IT will ask about anyway.
Answered here so the questionnaire is a formality, not a discovery exercise.
01
Where the data actually sits
The application, the database and the automation engine run in Warsaw, on OVH infrastructure inside the EU. Model calls go to Anthropic's first-party API and are processed outside the EEA, under the EU Standard Contractual Clauses: identifiers are masked before the call, and the provider does not train on the content. An EU-resident endpoint on AWS Bedrock (Frankfurt) is available on request. We separate hosting from inference because your security review will.
02
DPA available for signature
A data-processing agreement with SCCs for any non-EEA transfer, a maintained subprocessor list with 30-day notice of change, support for data-subject requests, and a 72-hour breach-notification commitment.
03
Masked before the call
Account numbers, national identifiers and card numbers are tokenised before any model call and restored after – the model never sees them raw. Under its commercial terms the provider does not train on the content.
04
A log nobody can edit
Authentication, role changes, every run, every approval and every export land in an append-only log, tenant-scoped and retained at least twelve months. It is a record of what happened, not a report assembled afterwards.
05
For supervised financial firms
We are prepared to contract under the MiFID II outsourcing regime (Reg. (EU) 2017/565) and to supply DORA register data, incident-reporting deadlines, audit rights and a defined exit – with no sub-outsourcing without your consent. Readiness, stated as readiness: we claim no certification we have not obtained.
06
Isolation and encryption
Tenant isolation enforced by PostgreSQL row-level security rather than application checks alone, sessions bound to your own subdomain, TLS 1.2+ in transit and AES-256 for stored credentials.
The full security overview and the subprocessor list are available on request, and before signature.
Direct questions, direct answers.
Where is the data processed?
Your workspace, database and automation engine run in Warsaw, on European infrastructure. Backups are encrypted and held in EU-jurisdiction storage. Model calls go to Anthropic's first-party API and are processed outside the EEA, under the EU Standard Contractual Clauses; identifiers are masked beforehand, and the provider does not train on the content. If your policy requires inference inside the EU, we route calls to AWS Bedrock in Frankfurt under a separate arrangement.
Does the model train on our data?
No. Identifiers are masked before the call, and the model provider is bound by no-training terms. We record token counts, cost and model, never the content of a prompt or a response.
We are a supervised financial firm. Is this outsourcing?
It may be, and we are set up for it. We are prepared to contract under the MiFID II outsourcing regime (Reg. (EU) 2017/565) with audit rights, a defined exit, and no sub-outsourcing without your consent. How the arrangement is classified is your call and your counsel's – we supply what that assessment needs.
Does the log replace our statutory records?
No. It is an operational trail of what Keelra did and who approved it. Your statutory record-keeping obligations remain yours, and the trail is built to feed them rather than to stand in for them.
Twenty minutes and one process to name.
No slide deck. You leave the call with a named process, an estimated gain, and a pilot quote.