# Security: the strongest control we offer is not receiving your data

How Tileward is secured. Choose the data boundary first: Tileward can run as a hosted service, inside your VPC, or fully air-gapped, and in the last two modes the inference data path stays on infrastructure you control. The page then covers the governance controls, audit retention and export, certifications, and how to report a vulnerability.

Page: https://tileward.com/security/

## Pick your exposure

- Hosted API: we hold prompts, context and audit records, and everything on this page applies. Zero infrastructure for you.
- Your VPC: one node, one GPU. Content, context and audit stay in your account, under your controls and your network boundary.
- On-prem, air-gapped: no network path out, a flat licence with no metering, and audit records retained locally.
- If your threat model does not permit prompts leaving your network, choose air-gapped. [See the complete request path](https://tileward.com/platform/).

## Governance as a control

- A request on a locked tile is classified and declined before inference (roughly 14 tokens to decide, and no completion written).
- Policy travels with the credential: rules are bound to the API key, not to a system prompt. A caller cannot drop the policy by changing the prompt, and a leaked key carries only the permissions attached to that key.
- Every decision is recorded, allow and refuse alike: the time, the decision, the tiles it matched, how the match was made, the size of the request, the API key or access token that made it, the actor your client names in its request header, and, where they apply, the policy version in force, the invoking model and the target model. Exportable as CSV or JSON on every account; webhook and SIEM export are Enterprise.
- Matching is by tile, not by keyword: a reworded request in the same vocabulary lands on the same tile, while wording that avoids the tile's vocabulary altogether can still pass.

More on [Tileward Governance](https://tileward.com/governance/).

## Audit retention

| Account | Audit window | Export |
|---|---|---|
| Pay as you go | 7 days | CSV, JSON |
| Enterprise, self-hosted | Local, your policy | CSV, JSON, webhook, SIEM |

For records kept longer, the self-hosted route is the one where retention is yours to set.

## Certifications

- SOC 2 Type II: in progress, monitored by Comp AI.
- ISO 27001: in progress.

Details: https://tileward.com/security/#certifications

## Reporting a vulnerability

Send it to [hello@tileward.com](mailto:hello@tileward.com) and it will reach an engineer. Please give us a chance to fix it before publishing.

The full text is at https://tileward.com/security/.

## Related

- [Platform](https://tileward.com/platform/): the complete request path
- [Tileward Governance](https://tileward.com/governance/): tile locks before inference and the audit record
- [Privacy policy](https://tileward.com/privacy/): what Tileward holds and for how long
- [Data processing addendum](https://tileward.com/dpa/): the processor terms for the hosted API
- [Sub-processors](https://tileward.com/subprocessors/): each vendor and the data it sees
- [Terms of service](https://tileward.com/terms/): the agreement, in plain words