Capability · Framework — observability

Lilypad (Mirascope Labs)

Lilypad is a developer-first, open-source platform for the 'prompt-as-function' workflow. Every Python function you decorate with `@lilypad.trace` becomes a versioned, traced experiment — with inputs, outputs, model config, and costs captured. Lilypad pairs naturally with Mirascope's prompt DSL but works standalone.

Framework facts

Category: observability
Language: Python
License: MIT
Repository: https://github.com/Mirascope/lilypad

Install

pip install lilypad-sdk
# run the self-hosted server (docker) or use hosted cloud:
docker run -p 8000:8000 mirascope/lilypad

Quickstart

import lilypad
from openai import OpenAI

lilypad.configure(project='my-app', base_url='http://localhost:8000')

@lilypad.trace(versioning='automatic')
def summarize(text: str) -> str:
    resp = OpenAI().chat.completions.create(
        model='gpt-4o-mini',
        messages=[{'role':'user','content': f'Summarise: {text}'}]
    )
    return resp.choices[0].message.content

summarize('The mitochondria is the powerhouse of the cell.')

Alternatives

Arize Phoenix — larger OSS peer
Langfuse — self-hostable alternative
Braintrust — hosted prompt + eval platform
W&B Weave — eval-oriented

Frequently asked questions

Do I need Mirascope to use Lilypad?

No — Lilypad works with any Python function that calls an LLM. Mirascope just makes the prompt DSL nicer to trace.

What does 'versioning=automatic' mean?

Lilypad snapshots the function's source, model config, and inputs each call, so the same function changes over time are tracked as separate versions with their own eval histories.

Sources

Lilypad — docs — accessed 2026-04-20
Lilypad GitHub — accessed 2026-04-20