Examples
Every notebook has committed outputs and an Open in Colab badge, and runs offline: mock LLM, scripted agent models, mocked HTTP. A few use local Docker containers (Postgres, Vault, Jaeger, Tempo + Grafana); with PECCA_SKIP_DOCKER=1, which CI sets, those cells are skipped and a fallback runs. CI executes every notebook with PECCA_TEST_SMALL=1. The Colab badges work for notebooks that don't need Docker.
| Notebook | What it shows | Runtime | Needs Docker? |
|---|---|---|---|
| Quickstart | connect → profile → train → shadow → status → audit pack | ~3 min | no |
| LangGraph | Pecca as the agent's first tool | ~15 s | no |
| Strands Agents | same, with Strands | ~15 s | no |
| Google ADK | same, with ADK | ~15 s | no |
| OpenAI Agents SDK | same, with the Agents SDK | ~15 s | no |
| Claude Agent SDK | same, through the SDK's MCP server | ~30 s | no |
| Connector: data sources | CSV, Parquet, Postgres (container) and OTel traces read into one schema | ~10 s | yes |
| Connector: registry | the same call on the local and mlflow registries |
~35 s | no |
| Connector: notifiers | webhook, Slack and Teams payloads; events firing in a project | ~15 s | no |
| Connector: ticketers and the approval loop | full approval loop with manual, GitHub Issues and Jira | ~25 s | no |
| Connector: schedulers | cron, pecca scheduler run --once, Airflow DAG, Databricks job JSON |
~20 s | no |
| Connector: serving targets | serving latency (pickle vs ONNX) and the docker_export folder |
~15 s | no |
| Connector: secrets | env, HashiCorp Vault (container) and AWS Secrets Manager (mocked) | ~10 s | yes |
| Connector: docs publishers | Markdown folder and Confluence (mocked) publishing | ~15 s | no |
| Monitoring with Jaeger | OTLP spans into a local Jaeger, read back via its API | ~20 s | yes |
| Monitoring with Grafana Tempo | spans in Tempo, TraceQL queries, provisioned Grafana dashboard | ~90 s | yes |
| Monitoring with MLflow | what MLflow shows for Pecca versions, state and logs | ~40 s | no |
Monitoring
The Jaeger, Grafana Tempo and MLflow notebooks above run real stacks from examples/monitoring/ (Docker Compose). For other vendors Pecca has no vendor-specific code: these are config-only guides:
Datadog · New Relic · Honeycomb · Langfuse · Dynatrace.
See also OpenTelemetry.