Shadow-AI detection, outbound spend tracking, AI Bill of Materials, MCP fleet monitoring — the four artefacts the EU AI Act and SOC 2 reviewers will ask for, generated continuously and exportable.
Discover every LLM runtime, model file, framework and CUDA stack on your fleet.
Watch outbound traffic to AI APIs. Estimate cost, attribute it, set budgets.
EU AI Act risk-classified inventory. PDF + CycloneDX 1.5 export.
Auto-discover MCP servers. Health, tools, tool-injection alerts.
The agent scans every host every five minutes for LLM runtimes, model files, AI Python packages, and CUDA stacks. New finds become events; whitelisted ones go quiet automatically.
The agent watches connections to known AI hosts — OpenAI, Anthropic, Azure OpenAI, Bedrock, Gemini, Groq, Mistral, Cohere, Perplexity, Together, more. Cost estimated from a configurable pricing table; per-tenant overrides for Enterprise contracts.
Compliance-ready inventory aggregated from Shadow-AI detections. De-duped across hosts (the same framework on 10 servers is one BOM entry). Per-entry human metadata: risk class, purpose, data inputs, license, publisher.
MCP servers are the new attack surface — they own actions on your infrastructure. We auto-discover them via process signatures and localhost MCP initialize handshakes; manual registration covers external endpoints.
Continuous, auditable, exportable. Without an extra vendor.