146 liveAI Agents.
94 security agents and 52 AI payment agents, all in one subscription. Each one is documented in OpenAPI + MCP and reachable as a typed POST endpoint.
Live means a real typed endpoint that returns a model-backed result right now — not a mockup, not a waitlist. Every agent in this catalogue has one; none are marked in development. Each response shape is published at /openapi.json.
Live · Shadow marks the 2 active-defense agents (#145, #146). They are fully testable and return a complete containment decision carrying the exact payload an enforcement point would run — and they refuse to run it. See the note below.
In development, and not claimed as working anywhere on this site: connectors that push #145 and #146 decisions into EDR and network firewalls, and the non-crypto payment rail (cards / ACH), which returns 501/503 rather than a fabricated success. Crypto settlement works end to end on devnet.
shadow_mode: true until you connect an enforcement point and enable it — you see what would have fired, and a person or an integration is still the thing that fires it. For web and API traffic that enforcement point exists: the Runtime Shield runs the WAF rules #146 writes, once a person approves them. The aim is a balance between guardrails (safety, alignment, and compliance) and independent execution (agentic AI that decides without constant human prompts). On the roadmap: a strict execution sandbox for autonomous agents. It is a Phase 3 capability and is not built today, which is why the pair above ship in shadow mode.