← Build Analytics & AI

Hermetiq Data Sources

What Hermetiq collects, how it is stored, and how the evidence reaches Barn Owl.

Start with Bazel build events. Connect profiles, cache and execution records, and infrastructure telemetry for deeper analysis. Available analytics and AI investigations depend on your connected sources and permissions.

What we collect and how it connects

Build and invocation identity links the available targets, actions, tests, profiles, and cache records. Worker and infrastructure context helps relate the workload to the environment that ran it.

Bazel build data

Build Event Protocol (BEP)
Bazel streams build events to Hermetiq: targets, tests, failures, flags, and build metadata.
JSON trace profiles
Hermetiq retrieves uploaded profiles from your configured Content Addressable Storage (CAS) for build phases, action timing, parallelism, and resource counters.
CI change-impact snapshots
Optional bazel-diff snapshots uploaded by CI connect code changes to affected targets and test coverage.

Buildbarn action, cache, and configuration data

Action Cache lookup events
Buildbarn sends lookup events so Hermetiq can link cache hits and misses to individual actions.
Completed-action logs
Buildbarn workers send execution records: queue wait, input fetch, execution, output upload, results, and resource usage.
Content Addressable Storage (CAS)
With configured read access, Hermetiq retrieves action commands, input metadata, environments, and platform details to explain cache behavior.
Buildbarn configuration
Loaded ConfigSets and authorized configuration sources give Barn Owl the settings and files it needs to explain your configuration.

Infrastructure telemetry and costs

Metrics via VictoriaMetrics
Your observability pipeline collects Buildbarn and Kubernetes metrics; Hermetiq queries them for performance and health.
Logs and events via VictoriaLogs
Forwarded pod logs and Kubernetes events give Hermetiq operational context.
Kubernetes API
Authorized reads supply worker pools, workload configuration, replicas, and autoscaling state.
OpenCost
Hermetiq reads Kubernetes resource allocations and infrastructure costs from your configured OpenCost API.

Where the data is stored

Hermetiq uses different stores for build records, artifacts, and infrastructure telemetry. The services available depend on your Cloud or on-prem deployment.

PostgreSQL
Structured builds, invocations, targets, actions, tests, cache events, metadata, and parsed profile metrics. High-volume tables use time partitions for retention.
Cloud object storage
Raw trace profiles and optional compressed build logs use Google Cloud Storage or Azure Blob Storage. Build logs otherwise default to PostgreSQL with two-day retention; object-storage logs follow the bucket’s lifecycle policy.
VictoriaMetrics and VictoriaLogs
Infrastructure metrics stay in VictoriaMetrics. Pod logs and Kubernetes events stay in VictoriaLogs, where Hermetiq queries them when needed.
BigQuery, when configured
Selected PostgreSQL analytics records replicate to BigQuery to support analysis over longer history.

NATS JetStream buffers incoming build events while ingestion workers process them. Redis/Dragonfly caches query results and temporary application state.

Where cost data comes from

OpenCost supplies infrastructure costs and resource allocations for CPU, memory, GPU, storage, and network. Per-action cost estimates combine execution duration with worker/node pricing available through VictoriaMetrics.

These are allocated costs and estimates, rather than direct cloud billing invoices. Cost reporting requires the relevant integrations, pricing, and attribution data; build events alone do not supply it.

How the evidence reaches the AI

Barn Owl calls Hermetiq’s tools to retrieve build records, configuration, and infrastructure evidence available to the authorized project. The tool results become context sent to the configured AI model for its answer.

Which investigations it can perform depends on the connected sources, tool availability, and permissions. Compatible coding agents can query available project evidence through Hermetiq’s MCP integration.

Read about AI agents and data access →

Connect the sources you need

Your project’s Quickstart supplies the Bazel configuration. The documentation covers cache, action, profile, infrastructure, and cost integration prerequisites.