The first question is always “where does our data go?”
It's the right question. Most AI vendors answer it with a compliance badge and a link to a sub-processor list. Here's the direct version — what we do with your data, what we don't, and what you keep when you leave.
| Dimension | Policy |
|---|---|
| Training on your data | NO |
| Data residency | US-HOSTED |
| Encryption in transit | TLS 1.3 |
| Encryption at rest | AES-256 |
| Access model | RBAC |
| Audit logging | IMMUTABLE |
| Code ownership | YOURS |
| Exit package | 5 DAYS |
Six non-negotiable security controls.
Built into every architecture, deployment, and ongoing managed service.
Zero model training
Your proprietary data and client records are never used to train public foundation models.
End-to-end encryption
Data in transit and at rest is secured to AES-256 and TLS 1.3 standards.
Human approval gates
High-stakes outputs require sign-off before external delivery or financial sync.
Role-based access (RBAC)
Granular permissions ensure people only reach data their credentials allow.
Audit logging and telemetry
Complete immutable logs of every prompt, input, system update, and execution timestamp.
Zero vendor lock-in
All code, prompts, and integrations are owned and exportable by your business.
Our own site assistant
Conversations with the assistant on this site are logged for 90 days to improve our content, then deleted. We store a hashed IP rather than the address itself, we don't use conversations to train any model, and processing runs on a billed API tier with no data-sharing clause. Full detail in our privacy policy.
How failures behave.
Engineered for graceful degradation, explicit exception queues, and zero silent failures.
Failing safe by default
Every automated decision carries a confidence score. Below the threshold you set, the item stops and waits for a person rather than proceeding on a guess.
- Per-workflow threshold configuration
- Exception queue with stated reason
- Visible exception rate in reporting
Consequential actions need a person
Anything with external or financial consequence — a customer email, a payment, a contract change — requires explicit approval regardless of confidence.
- Configurable per action type
- Approver identity logged
- No silent auto-approval
No answer without a source
Knowledge systems answer from retrieved passages and cite them. Questions outside your documents get an explicit refusal, not an invention.
- Source document and section shown
- Out-of-scope refusal by design
- Coverage measured in reporting
Broken means visible
A pipeline that fails routes to a human queue and raises an alert. It never fails quietly, which is the failure mode that erodes trust fastest.
- Alerting before user impact
- Failed payload retained
- Full retry history
You can always go back
Nothing we build removes your ability to do the work manually. Automation runs alongside your process during pilot and can be switched off without stranding you.
- Parallel running during pilot
- Manual path preserved
- Documented rollback procedure
What you own.
Ownership, at every stage.
| Asset / Dimension | During build | After launch | If you leave |
|---|---|---|---|
| Source code | Yours | Yours | Yours |
| Prompts and configuration | Yours | Yours | Yours |
| Documentation | Delivered progressively | Complete | Complete |
| Your data | Never used for training | Never used for training | Returned or deleted on request |
| Hosting accounts | In your name | In your name | Remain yours |
| Exit package | — | — | Within 5 business days |
Working with your security team.
From vendor questionnaire to continuous audit logging and least-privilege credentialing.
1. Questionnaire
We complete your standard security questionnaire during scoping rather than after contracting.
2. Data Map
A written record of what data the system touches, where it goes, and who can reach it.
3. Access Review
Least-privilege by default. We ask for the minimum access the build needs and document every credential.
4. Approval Design
Your team decides which actions require a human gate. We build to that, not to a default.
5. Ongoing Audit
Access, logs, and permissions reviewed as part of monthly Managed AI → reporting.
Frequently asked questions.
Direct, verifiable answers to enterprise data privacy and security questions.
Send us your security questions.
Direct answers from an engineer who builds the pipelines — no generic compliance brochures.