Guide
Deploy your first agent
Create a Metrune project, define an environment, and ship a versioned agent release with production telemetry attached.
Get a production signal in minutes.
Deploy a service, connect an existing runtime, and follow your first trace through models, tools, and infrastructure.
Guide
Connect an existing runtime
Add Metrune to an API, worker, or Kubernetes service without changing how the workload is hosted.
Guide
Trace model and tool calls
Build a readable run timeline that connects provider requests, tool execution, retrieval, and application outcomes.
Guide
Configure budgets and reliability alerts
Turn production signals into actionable policies for spend, latency, failures, and agent outcomes.
Frequently asked
Technical answers, without the hand-waving.
Do we have to move our agents onto Metrune?
No. Instrument existing services with OpenTelemetry or the Metrune SDK, or deploy new workloads directly. Both paths appear in the same production view.
Which model providers are supported?
Metrune is provider-neutral and supports OpenAI, Anthropic, Google AI, Azure OpenAI, AWS Bedrock, and custom OpenAI-compatible endpoints.
How is token cost calculated?
Usage is captured per model call and attributed to agents, releases, environments, customers, and teams using current provider pricing or your negotiated rates.
Can we separate staging and production data?
Yes. Environments have separate credentials, retention policies, budgets, alert rules, and access controls while remaining comparable within one workspace.
Does Metrune store prompts and responses?
Payload capture is configurable. Teams can redact fields, disable body storage, set retention windows, or keep telemetry inside a private deployment.
What happens when a release regresses?
Metrune correlates reliability and cost changes with deployment history, alerts the owning team, and preserves the prior release for a fast rollback.