Trust
BVECrew crew members are AI staff, supervised by people. Every one of them says so when asked. This page is the plain-language summary of how the platform is put together, what it will and will not do with your data, and where we stand on certification. Nothing here is aspirational; if it is not built yet, it says so.
BVECrew runs as a set of internal services (the client connector, Anchor; the client portal, Front Office; the agency console, Back Office; and the crew runtime behind them) plus, for Vault placements, a physical appliance called Strongbox. Four properties hold across all of it.
01 / Transport
Services authenticate each other with short-lived certificates issued by our internal OpenBao PKI. No shared static secrets between services, no plaintext hop, and a certificate that expires quickly enough that a stolen one is close to worthless.
02 / Redaction
When a placement is allowed to use a cloud model at all, context is redacted on our side, on the same host, before it leaves. We never send your data to a third-party DLP or redaction API to have it cleaned. Confidential-policy and Vault-residency placements never call a cloud model.
03 / Audit
Every action a crew member takes, every approval a person gives, and every access to a client's data is written to an append-only log where each entry carries the hash of the one before it. Alter or delete a record and the chain breaks visibly. You can export your placement's log.
04 / Approval
Each action a role can take is classed R0 to R3 in its charter. The runtime, not the model, decides whether an action runs, is logged, or waits for a person. Your overlay can only tighten it.
Every crew member carries a maximum risk class on its badge. Anything above that class is not available to it at all. Within its class, this is what each tier means.
| Class | Meaning | Who acts | Example |
|---|---|---|---|
| R0 | Read-only | Runs unattended. Logged. | Reading a ticket queue, checking a dashboard, drafting a summary for a person. |
| R1 | Reversible internal writes | Runs unattended. Logged and reversible. | Updating an internal ticket field, filing a note, moving an item between internal queues. |
| R2 | External output, sent after human approval | Queued for a person. Sent only after approval. | An email to a customer, a reply on a public channel, an invoice reminder. |
| R3 | Irreversible / production, always human-approved | Always approved by a person. Two-person rule available. | A refund, a production change, a filing, anything that cannot be undone. |
Some roles also require client approval before specific actions regardless of class. Those are listed as "checks first" on each role's sheet on the crew page.
Residency is where your data and your crew member live. Three levels, and every placement picks one on the order form.
Read the full residency comparison. Under Vault residency the Strongbox appliance is on your premises and nothing leaves your building: no cloud model, no remote storage, no telemetry beyond what you allow.
Every BVECrew staff member is AI-operated and supervised by a person, and it says so. It identifies itself as an AI staff member in its profile, in the signature of anything it sends, and in plain words whenever anyone asks. It does not claim to be a person and it is never configured to. A client cannot switch this off in an overlay.
The full statement, including what to expect when you interact with one, is at /legal/ai-disclosure.
The complete list. We keep it short on purpose.
| Subprocessor | What for | Applies to |
|---|---|---|
| Cloudflare | DNS, CDN, Access (identity-gated portals) and Tunnel (no inbound ports on our hosts). | Platform edge. Does not process placement data. |
| Your chosen model provider | Language-model inference on context we have already redacted locally. | Standard-policy placements only. Never Confidential-policy placements, never Vault residency. You pick the provider on the order form; changing it is a subprocessor change with notice. |
Your own cloud account (Tenant residency) and your own premises (Vault, Keyed connector host) are your infrastructure, not our subprocessors.
Each client gets a dedicated encryption key. Everything we hold for you, including backups, is encrypted under it. At contract end, after your export window, we destroy that key. Without it the data is permanently unrecoverable, everywhere it was ever copied. This is crypto-shredding, and it is how we can delete backups without hunting for every one.
We then give you a signed deletion certificate identifying the placement, the key identifiers destroyed, when, and by whom. It is your proof for your own auditors and regulators. For Vault placements the on-site shred runs before the appliance leaves your building.
Vertical compliance packs (tax, financial services, retail) map controls to the regulations those buyers answer to. They are control sets, not certifications.
If you find a security issue in BVECrew or any Black Vault Engineering Group LLC service, tell us before you tell anyone else and we will work it with you. Good-faith research within scope will not be met with legal action.
[CONTACT: pending -- [email protected]] (placeholder -- HUMAN_ACTIONS: confirm real security contact address)
Include what you found, how to reproduce it, and how to reach you. We acknowledge reports within a stated window and keep you informed until it is fixed. [STATUS: pending -- acknowledgement window] (placeholder -- HUMAN_ACTIONS: set the acknowledgement SLA for vulnerability reports)