Agent Lens functional flow

Agent Lens functional flow A data-flow diagram generated by Archify. 01 / Emit 02 / Collect 03 / Stream 04 / Process 05 / Investigate Instrumented agent · agentlens.init() · 01 / Emit · OTLP + OpenInference Instrumented agent agentlens.init() OTLP + OpenInference OTel collectors · edge → platform · 02 / Collect · redact + identity OTel collectors edge → platform redact + identity Redpanda · Kafka API · 03 / Stream · otlp.traces.v1 Redpanda Kafka API otlp.traces.v1 Normalize · hashed spans · 04 / Process · worker Normalize hashed spans worker Assemble · one execution · 04 / Process · worker Assemble one execution worker Score · DM / Hamb / Vtraj · 04 / Process · async Score DM / Hamb / Vtraj async ClickHouse · hash-only durable · 05 / Investigate · tenant scoped ClickHouse hash-only durable tenant scoped API + dashboard · FastAPI → Next.js · 05 / Investigate · :18000 / :3000 API + dashboard FastAPI → Next.js :18000 / :3000 OTLP traces redacted traces raw traces spans.transient executions.score hashed spans scores + findings tenant query Legend primary data policy / PII data store data flow

Collect

  • • Edge is the only normal ingress
  • • Redaction and identity happen before the stream

Process

  • • Workers run off the request path
  • • Deterministic scores; no LLM-as-a-judge

Investigate

  • • ClickHouse is hash-only and tenant scoped
  • • Dashboard reads the API, not the store