Frequently asked questions.

Hermetiq helps teams continuously improve the speed, reliability, and efficiency of their Bazel builds by connecting build failures, cache behavior, and infrastructure performance to fixes they can review and verify. We offer managed remote cache and execution, or observability and Barn Owl, our AI build engineer, for the Buildbarn you already run.

Bazel and Buildbarn teams that spend too much time investigating failures, slow builds, cache misses, or infrastructure problems. You can keep operational control and get better tools for investigation, or hand off the agreed build infrastructure to Hermetiq.

Barn Owl is Hermetiq’s AI build engineer. It uses connected build and infrastructure evidence to investigate problems, explain regressions, and recommend fixes. Authorized coding agents can apply changes and rerun builds to validate the result. Unattended build monitoring and automatic fix PRs are coming soon.

Both. Managed builds provides remote cache and execution with build intelligence; intelligence for existing Buildbarn adds observability and AI while you keep operating your infrastructure. Both have cloud and on-prem options. Hermetiq operates Cloud Managed Builds; operating responsibilities for on-prem managed builds are agreed with you.

Bazel still builds and tests your code, and your CI still coordinates your pipeline. If you already run Buildbarn, you can keep it and connect Hermetiq; if you want managed builds, we agree which infrastructure and operations Hermetiq will take on.

Start with who will operate your builds, where the system must run, and how your team will investigate problems. Hermetiq’s proposition is managed builds or intelligence for your existing Buildbarn, with observability and Barn Owl in either path; evaluate the options against a real workload and your deployment requirements.

A connected view of build and infrastructure evidence, plus AI-assisted investigation. With the relevant telemetry connected, you can investigate slow invocations alongside action timing, queue pressure, worker behavior, and storage signals. The depth of the diagnosis depends on the records available for that build.

Grafana and logs remain useful for monitoring and investigation. Hermetiq relates available infrastructure signals to Bazel invocations, targets, actions, and build history, giving engineers and agents build-specific context for their investigation.

Coding agents can already debug code and call external tools. Hermetiq gives them structured build history and connected cache, execution, and infrastructure evidence through the Model Context Protocol (MCP), so they can investigate your build system with context beyond the current workspace.

You can connect the Buildbarn you already operate. Initial integration configures build-event reporting and the additional telemetry you want to use; changes to build rules or infrastructure may follow when an investigation identifies a specific improvement.

Start with Bazel’s Build Event Protocol (BEP) for invocation, target, test, and outcome data. Add JSON profiles for critical-path analysis, cache records for cache investigation, and completed-action and infrastructure telemetry for remote-execution diagnosis. Each view depends on the corresponding data being connected.

Yes. Hermetiq’s MCP integration supports clients including Claude Code, Claude Desktop, Cursor, Windsurf, and Codex. Available queries depend on your project’s telemetry, permissions, and server configuration; repository edits and terminal commands use the separate access you grant your coding client.

With the relevant profiles and remote-action telemetry, Hermetiq helps separate analysis, queueing, input transfer, execution, and output transfer, then compare equivalent builds. That helps you locate a regression and inspect the evidence before making changes.

With the relevant cache and action records connected, Hermetiq can compare inputs, commands, environments, and execution platforms across invocations. That helps distinguish expected invalidation from avoidable differences; an aggregate hit rate alone does not explain a miss.

Yes, when the relevant telemetry and cost inputs are configured. Hermetiq connects build demand to queues, workers, storage, and resource use so you can investigate bottlenecks, wasted work, and capacity decisions.

Our published customer case study describes a frontier AI lab using Hermetiq for observability across more than two million builds per day on Buildbarn it operates itself. Our build-failure and queue-optimization walkthroughs are controlled local examples using Codex and Hermetiq MCP. They demonstrate specific diagnostic and repair workflows, not production benchmarks or guaranteed speedups.

That depends on what is limiting them. We recommend evaluating a representative workload, identifying a specific bottleneck, and comparing equivalent runs before and after a change. Useful measures include elapsed time, queue delay, expensive cache misses, infrastructure use, and investigation time.

Yes. Hermetiq supports cloud and customer-managed deployments for both managed builds and intelligence for existing Buildbarn. Hosting location and operating responsibility are separate choices, so we agree both as part of the deployment scope.

Build observability starts with telemetry, but build data can contain commands, file paths, logs, and outputs. Remote execution also processes the inputs required by each action, and an agent proposing code changes needs appropriate repository access. The data and access required depend on the workflow you enable.

Self-hosting Hermetiq does not by itself keep an external coding agent’s inference on-prem. MCP results enter the connected client’s context, and subsequent processing follows that client and model provider’s configuration. Review both the Hermetiq deployment and the AI client’s data path.

For hosted deployments, our Terms of Service permit use of customer data Hermetiq receives to improve the service, including AI-model training. Our Privacy Policy also permits indefinite retention and use of anonymized, aggregated service-usage data. For customer-hosted deployments, our Software License Agreement grants no right to use customer data for training. Your applicable agreement and any connected AI provider’s policies govern; these permissions do not establish that your data is currently being used for training.

Production changes remain subject to the permissions and review workflow you configure. Unattended monitoring and automatic fix PRs are coming soon; automatically applying and validating production changes is a separate roadmap step. Today, connected coding clients make authorized changes using their own repository and terminal access.

You can start the conversation before adopting Bazel. For managed builds, we assess your codebase, toolchains, dependencies, and migration effort; the agreed workload needs a supported Bazel build before Hermetiq can operate it.

Start with Hermetiq Cloud or a demo using a build problem your team recognizes. For an existing Buildbarn deployment, connect build events first and then the signals needed for that investigation. Contact us for pricing and an agreed scope; integration effort depends on your deployment and telemetry.