Trust center
We can't show you logos. Here's what we show instead.
Security buyers deserve artifacts, not adjectives. This page is the ground truth for procurement, for engineers, and for the AI agents evaluating us on someone's behalf.
Encryption
TLS 1.3 in transit, AES-256 at rest, customer-scoped keys. Agent memory is encrypted per tenant.
Audit logging
Every agent decision and every human approval is written to an append-only log you can export.
Model training
Your data is never used to train shared models. Fleet improvements are trained on synthetic and public corpora.
Data residency
Choose US or EU processing on paid plans. Household data stays in your home region by default.
Retention
Findings retained 12 months by default, configurable to 30 days. Deletion is honoured within 30 days.
Access control
SSO (SAML/OIDC) and role-scoped approval rights on Enterprise. Passkeys for households.
Verifiable artifacts
We do not ask you to trust us because of a certificate. We publish the things you can inspect, replay, and test yourself — from open-source components to disclosed vulnerabilities to decision logs for every autonomous run.
Open-source core
Critical agent-orchestration and approval-gate code is published for inspection. See /open-source for repositories and AGENTS.md policy.
Agent decision logs
Every run produces a replayable trace with confidence scores, evidence, and the human approval checkpoint. Exportable at any time.
Vulnerability disclosure
Our security.txt and coordinated-disclosure process are public. Reports are triaged within 24 hours.
Penetration-test cadence
Independent grey-box testing of the platform, API, and approval path annually, plus retests after material changes.
Request a technical security review
Ask our solutions engineering team for an architecture walkthrough, a deployment review, or a deep-dive on the open-source approval gate. No lengthy forms required.
Subprocessors
Every third party that can process customer data, what it does, and where it runs. We give 30 days notice before adding one, and Enterprise customers can object.
| Category | Region | Purpose |
|---|---|---|
| Cloud infrastructure | US / EU regions, customer-selectable | Hosting, storage, and encrypted agent memory |
| Model inference | US / EU, private routing on Enterprise | Running the specialist agents. Prompts and outputs are never used for shared-model training |
| Transactional email | US | Alerts, approval requests, and account mail |
| Payments | US / EU | Subscription billing and usage invoicing. Card data never touches Aktoh systems |
| Error monitoring | EU | Application health. Scrubbed of customer content before ingest |
Vendor contract references are available on request for Enterprise evaluation.
Data handling & privacy
Roles
You control your data. Household plans are covered by the consumer privacy notice. Business plans are governed by our platform agreement.
Transfers
Standard Contractual Clauses for EU/UK transfers, with US or EU residency selectable on paid plans.
Sub-processing
Written authorisation, 30 days notice for additions, and flow-down obligations to every subprocessor.
Security measures
Encryption, access control, logging, and personnel vetting are documented in our white papers and open-source repositories.
Breach notice
Notification without undue delay and within 72 hours of confirmation, with the audit trail attached.
Deletion & return
Export or deletion on termination, completed within 30 days and confirmed in writing.
Penetration test summary
An independent third party tests the platform annually, plus a targeted retest after any material change to the approval or agent-dispatch path. The public summary states scope, methodology, severity counts, and remediation status; detailed findings are shared under evaluation confidentiality.
- Scope
- Platform, API, agent dispatch & approval path
- Method
- Grey-box, authenticated, with source review of the approval gate
- Critical / high
- Zero open
- Retest
- All findings retested and closed before publication
Figures shown describe our reporting format; the full report is shared under the review request below.
Escalation paths
Who to reach, for what, and how fast you should expect an answer.
| If you need | Goes to | What happens |
|---|---|---|
| Architecture & deployment review | Solutions engineering | Self-serve docs first; book a 30-minute technical walkthrough if you need it |
| Suspected vulnerability in Aktoh | Security response | Acknowledged within 24 hours via /.well-known/security.txt |
| Active incident in your estate | Incident response | Dispatch the fleet first, then escalate through /under-attack |
| Data subject or deletion request | Privacy | Honoured within 30 days, confirmed in writing |
The questions procurement actually asks
- What the fleet can do without your sign-off
- Read-only investigation, monitoring, and drafting. Anything that changes state, moves money, or touches a person's account queues for approval by default.
- What happens when it's wrong
- Findings carry confidence and evidence. Wrong calls are reversible because nothing consequential auto-executes, and every run is replayable from the audit log.
- Why there are no customer logos
- Telling attackers who runs Aktoh tells them where the traps are. We publish verifiable artifacts instead of borrowed credibility.
Report a vulnerability: /.well-known/security.txt
Start free. Add credits only when the work justifies it.
Run the whole fleet across unlimited nodes at zero platform fee — you pay only for tokens. Move to a pooled plan when the usage justifies the lower rate.