A gateway, not an agent.

Intercept

Request received - Every outbound AI API call is routed through the CattleGrid gateway before reaching any external provider.

Inspect

Policy applied - Content is scanned in memory against your data classification rules — PII, financial data, trade secrets, custom-defined terms.

Pass

Clean request proceeds - No policy match. The request continues to the AI provider unmodified, in the same round trip.

Block, Redact, or Flag

Policy match found - Depending on the rule: the request is stopped before it leaves CattleGrid, the matched content is redacted and the request continues, or the request proceeds with an alert logged for review — whichever your policy specifies.

File Upload & Document Inspection.

The same block/redact/flag policy engine applied to file uploads, not just typed prompts. Covers PDF, Word (.doc/.docx), Excel (.xlsx), OpenDocument Text (.odt), and RTF. The file is extracted, inspected, and discarded — never stored, never written to disk.

Same zero-retention posture as prompt inspection, just applied to a wider surface.

Reversible Tokenisation.

Identifiers are swapped for tokens before the request leaves CattleGrid, so the AI provider only ever sees the token.

On the way back, real values are restored from an in-memory map that exists for that single round trip only, then destroyed — no persistent key, nothing to store or escrow.

This is pseudonymisation under UK GDPR Article 4(5): you remain controller of the data throughout, and CattleGrid never holds a re-identification key you didn't already have. Included in every tier with a monthly allowance — full detail on the Anonymise page.

Gateway latency

Sub-8 ms target. Inspection overhead per request, subject to your upstream AI provider's own response time.

Architecture type

API proxy gateway. Network-level interception. No endpoint agents. No application modification required.

Data retention

Zero on the gateway path. Inspected content exists in memory only, is never written to disc, and sensitive fragments from blocked requests are never stored. Conversation history, persistent memory and compliance review are separate features — off unless you switch them on, and retention is under your control when you do.

Deployment model

SaaS. On-premises deployment is possible on the Custom tier, following a conversation with us to assess what you actually need. It is not an off-the-shelf option.

Jurisdiction

European by design. CattleGrid is UK-registered and UK-owned, with no overseas parent. Production runs on Civo UK, with Civo Frankfurt providing resilience — and available as a primary EU location for organisations that need their data resident inside the EU. Wherever it sits, it stays under UK and EU law, and no extraterritorial regime reaches the governance layer. One legal framework, and nobody outside it able to move the goalposts or demand access to your data.

Tenant isolation

Enforced at the database level. Each organisation's configuration, rules and audit data are isolated by row-level security and per-tenant namespacing.

Data residency

Choose where it lives. UK by default, on Civo UK in London. Where an organisation needs its data resident in the EU, we can run it from our Frankfurt site instead — scoped with you, in the same way as an on-premises deployment. Either way it stays inside European law.

Getting started

Hosted on European infrastructure

1. Connect your AI provider API credentials to CattleGrid.

2. Enable pre-built detection rules or define your own.

3. Point your applications to the CattleGrid gateway endpoint.

4. Verify via the test interface. Go live.

On-premises - custom engagement

Within your own infrastructure

For organisations with data residency obligations, air-gap requirements, or a security architecture that requires internal deployment. Available on the Custom tier, scoped with us against what you actually need — not an off-the-shelf option.

1. Deploy within your existing network boundary.

2. Configure against your internal security policies.

3. Integrate with existing SIEM and alerting infrastructure.

4. No outbound data. Audit logs remain on-premises.

OpenAI (direct)

Direct API access. Consumer and enterprise tiers.

Anthropic (direct)

Claude API. Direct integration.

Google Gemini (direct)

Direct API access to Google's model family.

Mistral AI (direct)

European-domiciled provider, of particular relevance for EU data sovereignty requirements.

Plus any OpenAI-compatible API.

If your provider exposes a standard OpenAI-compatible endpoint, CattleGrid covers it without custom integration work.

Credentials, encrypted at rest

API credentials connecting to AI providers are encrypted at rest using AES-256-GCM. They are never exposed in logs or audit exports.

Tenant isolation, complete

Each organisation's configuration, rules and audit data are isolated by row-level security and per-tenant namespacing, enforced at the database level rather than by application logic.

UK and EU law only

CattleGrid is UK-registered, UK-owned and running on European infrastructure. For regulated firms the question 'who else could compel access to this?' has one short answer, and it is a European one. No overseas parent company, no foreign disclosure regime reaching through the governance layer. For UK firms that means UK law. For firms in the EU it means data that never leaves the legal framework you already report under — and, where residency is a requirement rather than a preference, a European home for it.

Audit exports, compliance-ready

Logs are exportable as evidence for your own audits and UK GDPR governance obligations. Nightly canonical export to your own S3-compatible bucket is available, so the evidence lives somewhere we cannot reach.

End users

Nothing, ordinarily. AI tools continue to function as normal. When a request is blocked, the user receives a clear notification explaining why. Nothing to install unless your organisation chooses to roll out the desktop client. No training required beyond awareness of the policy.

Administrators

Configure detection rules via the visual dashboard. Define what categories of data are governed and how. Manage team access and permissions. All administrative actions are logged.

Security teams

Real-time dashboard visibility across the organisation's AI usage. Immediate alerts on policy violations — via dashboard notification, email, or Slack. Integration with existing SIEM infrastructure. Full audit trail for incident investigation.

Compliance officers

Exportable reports for regulatory audits. Demonstration of data protection controls in operation. Pre-built templates for UK GDPR governance evidence. Configurable data retention policies.