// security · data flow · deployment models
The answers your security review is going to ask for
Forward this page to whoever runs your security review. Every answer below is the one we would give under a signed questionnaire, and none of them get better if you ask twice.
> egress --check operational_data
x export → public cloud: denied by policy
> model --inspect payload
schema names + column types
+ row data: not transmitted
// illustrative — not customer data
// how data moves
Four steps, and you can audit every one
The platform is built so that the trust question has a structural answer rather than a promise. Each step below is something your team can verify from your own infrastructure.
Gateways record to an ordered event log on hardware inside your facility. Nothing depends on a link being up at the moment the machine does something.
PostgreSQL and Parquet on infrastructure you own, queried by DuckDB. Open formats, so the record stays readable with or without us.
A transformation is generated as SQL and shown to you alongside a real-data preview. You approve explicitly. Nothing runs on an implicit yes.
Operational and administrative activity is recorded with least-privilege roles enforced at the data layer, and any window can be replayed.
// deployment models
Three shapes, one posture
Self-hosted on your own hardware is the default and the one we recommend. A private cloud tenant — your Azure subscription, private endpoints — is the middle option. A hosted single-tenant container exists for the self-serve pipelines tier, and it is isolated per customer rather than a shared database with a tenant column.
Whichever shape you pick, the operational data stays inside a boundary you define. Your data, your infrastructure, your configuration, and your deployment artifacts are yours permanently; the reusable platform core is licensed to you. If we disappeared tomorrow the deployment would keep running on your hardware and you would keep every artifact.
// the ceiling on what we claim
What we do not do
Engineered for ITAR, EAR, CMMC, and NIST environments. We are designed to support your compliance posture; we do not certify compliance. We are not an assessor, we do not issue attestations, and no deployment of ours makes an organization compliant on its own. What we do is build and operate the system so that the evidence your auditor wants is a byproduct of the work rather than a scramble at the end.
// faq
Security — straight answers
Where does our data live?
On your infrastructure. The operational record runs in PostgreSQL on your servers and the analytical tier is Parquet on storage you control. Where a hosted deployment is chosen instead, it runs in an isolated single-tenant container — never a shared multi-tenant database.
What reaches the AI model?
Your instruction, plus table and column names and their types. Row data does not reach the model. The model proposes a transformation as SQL; you read that SQL and a real-data preview, and nothing executes until you approve it explicitly.
What happens if The Mad Botter is down?
Your systems keep running. The record is written locally by design, so capture continues offline and syncs when the link returns. Downstream, a Power BI dataset keeps serving the last delivered data — an outage on our side delays a refresh, it does not empty a dashboard.
Can this run air-gapped?
Yes, and air-gap is a supported configuration rather than a workaround. The platform is offline-first everywhere: the edge writes an ordered event log locally and reconciles when a link exists. In a fully air-gapped facility there is no link, and the platform still works.
Who on your side can access our systems?
A named delivery lead and the engineers on their deployment, under credentials you issue and can revoke, with access scoped to the systems in the agreement. The team is 100% US-based; there is no offshore layer, no subcontractor pool, and no shared support account. Administrative activity is logged on your side, not just ours.
Do you certify our compliance?
No. Engineered for ITAR, EAR, CMMC, and NIST environments. We are designed to support your compliance posture; we do not certify compliance. Anyone who tells you their software makes you compliant is selling you something that does not exist.
If we stop paying, do we lose access to our own data?
No. You own your data, your infrastructure, your configuration, and your deployment artifacts. We retain the reusable platform core and license it to you; updates and managed operations run on a platform agreement. The record is written in open formats on hardware you control, so a lapsed agreement stops platform updates and managed operations — it does not stop the deployment and it cannot lock you out of your own data.
How do we report a vulnerability?
Email [email protected] with the details and we will acknowledge it. We would rather hear about a problem from you than from a customer, and we will tell affected customers ourselves.