monocle

Monocle is an open‑source framework that lets you add tracing to generative AI applications written in Python. It is presented as a tool for developers and platform engineers who need to watch their GenAI services while they run in production. The project lives on GitHub, and a Discord channel is offered for community questions and contributions.

The library's main job is to instrument code so that each request generates a trace that follows the open‑source cloud‑native observability standards, such as OpenTelemetry. In practice you import the Monocle package and wrap a function, often a model inference call or a chatbot handler, with a provided decorator. When the function executes, Monocle creates a span, records attributes like input prompt text, model name, and latency, and then closes the span when the response is returned. The output of the instrumentation is a trace record that can be sent to any back‑end that understands the OpenTelemetry format, for example Jaeger, Prometheus, or Grafana. No custom exporters are required; you configure the destination URL and authentication in a small settings file, and Monocle handles the rest.

The framework is aimed at teams that already have an observability stack and want to extend it to cover their generative AI workloads. If you are building a new model‑serving microservice, a chatbot, or a pipeline that chains multiple LLM calls, Monocle can capture the end‑to‑end flow without rewriting business logic. The quickstart guide on the site shows a minimal chatbot example: a single Python file imports Monocle, adds the decorator to the chat handler, and runs the app. When a user sends a message, the trace includes the incoming request, the prompt sent to the model, and the generated reply, all visible in the chosen tracing UI.

Because the code is open source, you can read how spans are built, modify the library to add custom attributes, or contribute fixes through pull requests. The documentation mentions a user guide that covers setup steps, exporter configuration, and naming conventions for spans. The Discord channel provides a place to ask about integration issues or share patterns. The site does not list any pricing because the software is free to use, and it does not specify support guarantees or service‑level agreements.

Key features