AI agents, analytics, applications and automations all want to. Cascade is the runtime layer that enforces your data-use policies across databases, applications and AI agents — making every request deterministic, inspectable, and backed by verifiable evidence.
Data went from stored, to distributed, to actively utilized — but the infrastructure that connects policy to actual system behavior never caught up. Rules exist. Systems act. Nothing between them is deterministic.
Customer PII may be used only for support and billing.
AI systems may not receive raw contact data.
Production health data never leaves the de-identified zone.
Inactive customer records are deleted after 24 months.
An AI support agent asks for a customer record. Here is everything Cascade does between the ask and the answer — in about sixteen milliseconds.
Cascade sits in the path between consumers and systems. A raw query, API call or agent tool-call is captured and normalised into who is asking, for what, and why.
The actor is verified against your identity provider, requested fields are matched to their classification, and the stated purpose is checked against the policy version in force.
Not just allow or deny. Cascade can mask, tokenise, filter rows, or route the request to a de-identified copy — the same answer every time for the same inputs.
Only the permitted shape of the request is executed, with Cascade-held credentials. The consumer never touches the data it wasn't entitled to.
What touched the data, which fields, for what purpose, under which policy, what control applied and what resulted — recorded per interaction, and verified.
Cascade doesn't need a separate product per consumer. Every use of data — human or machine, read or write — is expressed as an intent and answered by the same policy engine.
§4.1 — AI systems may not receive raw contact data
10 of 14 fields returned · 4 masked
The runtime layer that decides, controls and records how enterprise data is used — by every consumer, against every system, at the moment of use.
Data went from stored, to distributed, to actively used by software that acts on its own. Governance kept describing data. Nothing was built to govern its use. That is the gap this category fills.
Cascade complements what you own. It reads classifications from your catalog and DSPM, obligations from your privacy platform and identities from your IdP — then makes them enforceable everywhere data is used.
Cascade separates the power to propose from the power to act. Models plan; people and policy sign; a deterministic runtime executes inside your credential boundary; an independent observer checks what really happened.
Same product, same policy language, same guarantees. You choose where the control plane lives. Either way, the runtime that touches your data runs inside your network.
| Cloud | Self-hosted | |
|---|---|---|
| Control plane | Hosted and operated by Cascade | Runs in your environment |
| Runtime — the part that touches data | Your network | Your network |
| Crosses your boundary | Policies, decisions, trace metadata · outbound-only | Nothing |
| Row-level data leaves your network | Never | Never |
| Upgrades | Continuous, managed by us | Signed bundles, on your release cadence |
| Air-gapped operation | Supported | |
| Best for | Fast start · cloud-first estates | Regulated, sovereign or classified environments |
One place to see every decision, every system in scope, every drift from policy — and to sign off the few actions that need a human.
Cascade is data utilization infrastructure: a runtime layer that sits between data consumers (AI agents, analytics, applications, pipelines and people) and the systems that hold enterprise data. It decides, controls and records how data is used at the moment of use, based on your written data-use policies.
Data-use policy enforcement means making a written rule, such as “customer PII may be used only for support and billing”, actually determine what a system can do with data. Instead of reviewing access after the fact, each request is evaluated against policy before data moves, including who is asking, which data is involved and for what purpose.
An agent’s query, API call or tool call is intercepted and turned into a structured intent: the actor, the operation, the target data and the stated purpose. Cascade resolves identity against your identity provider, matches fields to their classification, and evaluates the policy version in force. It can then allow, mask, tokenise, filter rows, route to a de-identified copy or deny the request, and only the permitted shape of the request is executed.
Purpose-based access control grants or limits access according to why data is being used, not only who is using it. In Cascade, each request carries a declared purpose (for example customer_support or revenue_analysis) that is checked against the policy, so the same actor can be permitted for one purpose and masked or denied for another.
Cascade records each interaction: what touched the data, which fields, for what purpose, under which policy version, which control applied and what resulted. The resulting system state is independently verified, so teams can answer questions about data use from a decision record rather than reconstructing it from scattered logs.
Catalogs describe data, DSPM tools discover and classify sensitive data, privacy platforms document obligations, and access tools gate entry to a warehouse or database. Cascade is designed to complement them: it reads classifications, obligations and identities from those systems and makes them enforceable at the point where data is used, across consumers and systems.
The Cascade runtime, which holds credentials and touches data, runs inside your network in both deployment options. With Cascade Cloud, the control plane is hosted by Cascade and only policies, decisions and trace metadata cross your boundary, outbound-only. With Cascade Self-Hosted, every component runs in your environment, including air-gapped ones.
Cascade is built for security, platform and data engineering teams, and the CTOs and CISOs who own them, at enterprises and B2B SaaS companies that handle sensitive customer data and are putting AI agents, analytics and automation in front of it.
Cascade is engaged through pilots. A pilot runs against a controlled subset of your estate, with one consumer, one system and one policy, in observe-only mode first. Use the “Request a pilot” link on this page to get in touch.
Pilots run against a controlled subset of your estate — one consumer, one system, one policy — in observe-only mode first.