Governed AI-gateway telemetry

DOME SYSTEMSbuyer room · privateManage buyer room

Governed AI-gateway telemetry: a single normalised Audit v1 trail over every governed LLM/agent request, carrying allow / deny / filtered / failed outcomes, service and policy-path health, latency broken out per tool, model, gateway and rule evaluation, rate-limit and resource-exhausted quota events, and LLM consumption attributed by agent, API key, identity and served model across a six-class token partition (fresh input, cache-read, cache-write, output, hidden reasoning, call count) with per-class pricing applied at read time as estimated_cost_usd. Consumption is exposed as four read-time usage cubes plus time-series through documented REST endpoints (/api/v1/metrics/summary, /metrics/agents, /metrics/agents/{id}, /metrics/agents/{id}/keys), and usage history is explicitly retained after API-key revocation.