Shield Agent AI

← Docs hub

Customer guide

OTel verdict export

KYA works next to your observability stack, not instead of it. Tracing tools explain what the agent did; KYA decides what it may do and proves it. Verdict export puts that proof where your dashboards already live.

export KYA_OTLP_EXPORT_ENDPOINT=http://127.0.0.1:4318
# or in .kya/config.json:
# { "otlpExport": { "endpoint": "http://127.0.0.1:4318" } }
  • Every verdict becomes an OTLP/JSON GenAI span (gen_ai.tool.name, kya.verdict, kya.reason_code, token attributes when known).
  • Works with any OTLP/HTTP backend: Grafana, Datadog, Langfuse, a Collector, and friends.
  • Long-lived processes (MCP gate, gateway) export every verdict; hook / wrap spawns export DENY / REQUIRE_APPROVE via the detached flush helper.
  • Plaintext http is accepted for loopback only unless "insecure": true.

The gateway also runs a loopback-only OTLP traces receiver, so spans from other MCP servers become trail events. Full env list, Grafana / Datadog recipes, and a Collector sample: docs/otlp.md in the public repo; the positioning note is docs/marketing/observability-stack.md.

Hosted plane: richer Micrometer gauges and timers with KYA_OTLP_ENABLED=true.

Docs hub · Install · How KYA works · Source