candela-local
Developer proxy + runtime manager — run local & cloud models with observability from your machine.
Getting Started
Welcome to Candela! This guide will help you get up and running with LLM observability in minutes.
What is Candela?
Section titled “What is Candela?”Candela is an open-source observability platform that gives you full visibility into your LLM traffic. It consists of four components:
candela-server
Full backend — API server, LLM proxy, dashboard UI, deployed on Cloud Run.
candela-sidecar
Lightweight production proxy (< 5MB) for containers — traces, cost, budget enforcement.
candela-desktop
Flutter desktop app for managing providers and viewing traces.
Choose Your Path
Section titled “Choose Your Path”Use candela-local for instant LLM observability on your machine — local models via Ollama, cloud models via ADC, all traced to SQLite.
go install github.com/candelahq/candela/cmd/candela-local@latestDeploy candela-server on Cloud Run with BigQuery storage, Firebase auth, and a dashboard UI. Developers connect via candela-local in Team Mode.
Drop candela-sidecar next to your application containers for zero-touch LLM tracing in production.
Integrate Candela with your ADK agents for end-to-end distributed tracing with W3C Trace Context propagation.
Next Steps
Section titled “Next Steps” Installation Install any Candela component from the monorepo.
Quick Start Send your first traced LLM request in under 5 minutes.