# Everything your questionnaire asks, in one place.

**Everything your questionnaire asks, in one place.**

You have been sent here by a colleague who wants to buy this. Here is the control summary, what an audit export actually contains, how the platform behaves when it does not know, and the document pack — without a call.

The document portal is click-through NDA gated. Everything on this page is not.

For compliance & risk

[Request the security pack](/trust/docportal) [Open the Trust Centre](/trust/trust)

## Control summary.

Scannable, because you have eleven of these to get through this week. Each row links into the Trust Centre page that carries the detail and the evidence.

| Control | Position | Detail |
| - | - | - |
| **Tenant isolation** | Logical isolation with per-tenant key scoping; dedicated and VPC options for physical isolation. | [Security](/trust/security) |
| **Encryption** | In transit and at rest. Customer-managed keys on Cloud Dedicated and all Enterprise models. | [Security](/trust/security) |
| **Secret handling** | Encrypted secret references, injected at call time. Never in prompts, logs, manifests, exports or support tooling. | [Security](/trust/security) |
| **Access control** | RBAC with segregation of duties, OIDC and SAML SSO, enforced MFA. SCIM not yet shipped. | [Security](/trust/security) |
| **Data residency** | Singapore default. Region chosen at workspace creation and immutable thereafter. | [Residency](/trust/residency) |
| **Retention & deletion** | Configured per tenant per data class. Deletion is verified and evidenced, including from backups on the stated cycle. | [Privacy](/trust/privacy) |
| **Audit immutability** | Append-only records for the configured retention period. Export is itself an audited event. | [Governance](/governance) |
| **Model training** | No customer data is used to train models, ours or a provider’s. Provider zero-retention terms are a condition of the approved model list. | [Responsible AI](/trust/responsible) |
| **Privileged access** | Break-glass only, four-eyes approved, time-bound, fully recorded, and surfaced to the tenant. | [Security](/trust/security) |
| **DLP & redaction** | Configurable redaction on ingestion and on export, by data class, applied before model dispatch. | [Security](/trust/security) |
| **Incident response** | Defined severities, notification commitments in the DPA, post-incident reports published to affected tenants. | [Status](/trust/status) |
| **Subprocessors** | Published live table with entity, purpose, data categories and region. Change notification by subscription. | [Subprocessors](/trust/subprocessors) |

## What an audit actually gets.

Not a log file. A reconstructable account of a decision, scoped to the matter you asked about.

```
$ krewos audit export --matter SG-2291 --format json

run          r-8f21-4c  ·  2026-02-11T09:14:02+08:00  ·  2h 41m
release      4.2.0  ·  sha256:c41b9e7a…  ·  sealed 2026-02-03
components   41 pinned versions, each with checksum
model calls  17  ·  profile balanced-v3  ·  0 fallbacks fired
              provider jurisdictions: SG, SG, SG  (policy-constrained)
retrievals   41 chunks from 4 knowledge packs, each with
              source id, authority grade, date, relevance score
evidence     9 claims  ·  8 above floor  ·  1 abstained
policies     15 evaluated  ·  14 pass  ·  1 marginal (recorded)
actions      0 external effects  ·  no idempotency keys required
humans       1 approver, named · 2 edits with reasons
              · decision at 11:47  ·  elapsed in queue 2h 33m
autonomy     stage: regulated floor  ·  sampling: n/a (100%)
cost         412 platform + 289 model credits = S$7.01
              0 unattributed

export recorded as audit event a-9930 by user u-114
```

**Scoped, not wholesale**

Export by matter, by date range or by crew. An auditor asking about one matter does not receive every other client’s data, and the scope is recorded.

**Machine-readable and signed**

JSON and CSV for analysis, signed PDF for a bundle. The signature covers the record set, so tampering after export is detectable.

## How the platform behaves when it does not know.

Most AI vendor assessments have no question for this, because no vendor has previously offered an answer. It is the section worth reading twice.

**Abstention**

Below the declared evidence floor the crew returns a structured insufficiency result naming what could not be established and which sources would resolve it. It routes to a human knowledge-gap queue and is reported separately from failures.

**Conflict surfacing**

Where sources of comparable authority disagree, both positions are carried forward with their authority and dates. No silent selection, and the conflict is visible on the artefact and in the export.

**Staleness propagation**

When a source expires or materially changes, every artefact that relied on it is identified and its owners are notified. You find out from the platform, not from a client.

What this does not do: it does not make the model correct. An evidence floor governs whether there is enough support to proceed, not whether the reasoning over that support is sound. Evaluation gates and human approval carry that load, which is why regulated output classes stay at 100% review permanently.

## The document pack.

Click-through NDA, then immediate download. No sales call in between, and no form that routes to a queue.

**Data Processing Agreement**

Roles, processing purposes, subprocessor terms, transfer mechanism, breach notification commitments, deletion obligations.

**Service Level Agreement**

99.9% monthly Control Plane availability, exclusions, credit schedule, support response targets by severity.

**Penetration test summary**

Third-party annual test plus release-triggered testing. Summary letter with scope, methodology and remediation status.

**Security whitepaper**

Architecture, isolation model, secure SDLC, dependency and sandbox controls, privileged access, backup and disaster recovery.

**Pre-answered CAIQ and SIG-lite**

Completed, dated and maintained. Send us yours if it differs materially and we will complete it.

**BCP / DR summary**

RPO 15 minutes, RTO 4 hours on the standard tier, with the test cadence and the last test date.

**Insurance certificate**

Professional indemnity and cyber cover, current certificate.

**Subprocessor list**

Entity, purpose, data categories and region. Published openly with change notification by subscription.

### Certification status, stated honestly

We publish where each certification actually stands rather than implying more. If a control is not yet independently assured, the Trust Centre says so on the page for that control.

[Certifications & assurance →](/trust/certs)

### Book a compliance review

60 minutes with a solutions engineer and, where useful, our security lead. Bring your questionnaire. We will complete it live rather than promise it by Friday.

[Book a compliance review](/book-demo)

---

Canonical page: https://dev.krewos.ai/role-compliance