Observability Dashboard for Agentic Systems

A real-time observability service for multi-agent systems. It ingests OpenTelemetry traces over OTLP/HTTP, converts spans into structured agent events (lifecycle, LLM calls, tool invocations), and broadcasts them via WebSocket for live dashboards.

📖 Documentation: https://docs.agentic-layer.ai/observability-dashboard/

Development

Prerequisites

• Python 3.13+
• uv (Python package and venv manager)
• Docker
• pre-commit, tilt, kubectl (all installable via brew bundle)
• A local Kubernetes cluster (only needed for the Tilt path)

Build and run locally

# Install system dependencies
brew bundle
# Install Python dependencies
uv sync
# Install pre-commit hooks (one-time)
pre-commit install

# Hot-reloading dev server
make dev
# Or, production-like server
make run
# Or, run inside a local Kubernetes cluster
tilt up

The service listens on http://localhost:10005 (Tilt UI: http://localhost:10350).

Test

make check-fix  # linters and formatters with auto-fix
make check      # all checks and tests
make test       # tests only

Verify the local deploy

Confirm the service is up and accepting traces:

# Health probe
curl http://localhost:10005/health
# Subscribe to the global event stream
websocat ws://localhost:10005/ws

To generate traffic, point an agent's OTEL_EXPORTER_OTLP_ENDPOINT at http://localhost:10005/v1/traces.

Contributing

See the Contribution Guide. Documentation lives in docs/ (AsciiDoc, Diátaxis layout, built with Antora).