SOLUTIONS
One engine, every kind of traffic.
A chat request, a RAG context, a tool call, an MCP server or an outgoing email goes through the same inspection, gets one of the same five decisions and leaves the same kind of evidence. Each page below explains what is checked for one type of traffic, and in what order.
Protection
Prompt security
Injection, jailbreaks and hidden instructionsData protection
Masking personal data and secretsRAG and context
Document access, provenance stamps, indirect injectionAgent and tool security
Tool permissions, argument schemas, approval, risk budgetMCP security
Server inventory, tool descriptions, schema changesOutput protection
Leak, scope and brand checks inside the streamEmail protection
Body, HTML and attachment checks at an SMTP gateway
Platform
Policy engine
Five decisions, signed bundles, versions and diffsEvidence and audit
Position-based evidence, hash chain, WORM storageRed team
Campaigns, mutations, judging, findingsGateway and model library
OpenAI- and Anthropic-compatible endpointsDeployment: Cloud and on-prem
In your data centre or in a managed environment