Anvia

Observe systems ​

Anvia runtime events provide one consistent observation surface across providers, direct model calls, agent turns, and tools.

Log the lifecycle ​

ts
import { createConsoleLogger, createLoggerObserver } from '@anvia/logger'

const logger = createConsoleLogger({
  name: 'support-app',
  level: 'debug',
})

const agent = new Agent({
  id: 'support',
  model: model,
  observability: {
    observers: { logger: createLoggerObserver({ logger }) },
  },
})

Use local console logging while developing, then switch to the adapter for the logger your production application already operates.

What to watch ​

  • Run and turn start/finish events for latency and throughput.
  • Pipeline run and stage spans for orchestration latency and branch failures.
  • Tool calls and failures for integration health.
  • Final usage for model consumption.
  • Error events for failed runs.
  • Run and trace identifiers for correlation.
  • Runtime score distributions and end-user feedback trends for product quality.

Data safety ​

Do not treat observability as authorization. Verify access before an agent runs, and avoid logging message content or tool output unless your data policy explicitly permits it.

The default logger observer is conservative: it omits final output, full model requests and responses, and tool results.

For production traces that include Pipelines, pair the backend's Pipeline observer with its Agent observer and give both configurations the same primaryTrace name. This produces one trace from the Pipeline root through Agent generations and tools. @anvia/otel exposes createOtelPipelineObserver(); LensClient exposes pipelineObserver().

For feedback or automated production checks, record a trace-correlated runtime score with createOtelScorer() or LensClient.score(). A Boolean 1/0 can represent like/dislike, while numeric and categorical scores support richer product signals. Mark authenticated user-submitted feedback with source: 'end_user'; this provenance is observable metadata, not authorization.

Inspect locally ​

@anvia/studio provides a browser surface for running agents and inspecting sessions, traces, tools, memory, status, knowledge, pipelines, evaluations, and MCP configuration when those capabilities are enabled.

ts
import { Studio } from '@anvia/studio'

new Studio([agent]).start({ port: 4021 })

Open http://localhost:4021/playground and run a prompt against the configured agent.

Built for Anvia.