all projects

project 26 / 31

Self-Hosted LLM Observability

Arize Phoenix running locally (GCP) to trace LangGraph agent workflows - LangSmith-level insight into execution paths, latency and prompts, with sensitive data never leaving the machine.

#llmops#arize phoenix#langgraph#opentelemetry#observability#python#self-hosted#google cloud#gpu

About this project

Observability for LLM and multi-agent applications without sending data to a third-party SaaS.

CLI logs get messy fast when debugging complex agent workflows. Visualizing execution paths, latency, and exact prompt/response pairs is critical - but sensitive data can't always leave your infrastructure.

Self-hosted Arize Phoenix gives LangSmith-level insights locally: full trace trees of LangGraph chains, per-span status, inputs and outputs, and latency percentiles (P50/P99) across sessions. Easy to set up, full control over data security.

This setup runs on Google Cloud GPU infrastructure - training, tuning and serving included.

hosted on fly.io