loupe (Python SDK)

Instrument your LLM calls and ship latency, tokens, cost, and errors to a Loupe backend — in two lines.

# Not yet published to PyPI — install from this repo:
pip install ./sdk

import anthropic
from loupe import track

client = track(anthropic.Anthropic(), workload="support-bot")
# use `client` exactly as before — every call is now observed
resp = client.messages.create(
    model="claude-haiku-4-5",
    max_tokens=200,
    messages=[{"role": "user", "content": "hi"}],
)

Works the same for OpenAI:

from openai import OpenAI
client = track(OpenAI(), workload="rag-pipeline")
client.chat.completions.create(model="gpt-4o", messages=[...])

Configuration (env vars)

Create a key in the dashboard (🔑 API keys) or via POST /apikeys, then:

export LOUPE_API_KEY=loupe_sk_…

An API key is preferred over the admin login: it's scoped to ingestion, revocable per source, and needs no login round-trip. Or pass a configured Reporter:

from loupe import Reporter, track
client = track(anthropic.Anthropic(), workload="bot",
               reporter=Reporter(url="https://loupe.internal",
                                 api_key="loupe_sk_…"))

Notes

• Non-blocking & safe: reporting happens on a background thread and never raises into your code — if Loupe is down, your LLM calls are unaffected.
• Cost is computed by the backend from the model + token counts (you don't need to send it).
• v1 covers synchronous messages.create (Anthropic) and chat.completions.create (OpenAI). Async and streaming are on the roadmap.
• Pass provider="anthropic"|"openai" if auto-detection can't identify a client.