driftballoon 0.2.0

Python SDK for DriftBalloon - LLM output drift detection and observability

DriftBalloon Python SDK

LLM output drift detection and observability. One line of logging — DriftBalloon stays completely out of your critical path.

Install

pip install driftballoon

Quickstart

from driftballoon import DriftBalloon

db = DriftBalloon(api_key="db_sk_your_key")

# After each LLM call, log the response (fire-and-forget)
response = openai.chat.completions.create(model="gpt-4o", messages=[...])
db.log(
    name="support-agent",
    response=response.choices[0].message.content,
    model="gpt-4o",
).submit()

# Check which prompt version is active ("a" or "b")
active = db.get_active_prompt("support-agent")

Features

• Fire-and-forget logging — log().submit() is non-blocking; your app never waits on DriftBalloon
• Semantic drift detection — detects when LLM responses shift meaning or topic
• Length drift detection — catches abnormally short or long responses
• Prompt status tracking — check which prompt version is active via get_active_prompt()
• Multi-model support — track gpt-4o, claude-3-5-sonnet, or any model string
• Local config cache — prompt configs are cached and synced every 30s
• Offline resilience — your app keeps working if DriftBalloon is unreachable
• Retry with backoff — failed log submissions are retried automatically

Environment Variables

export DRIFTBALLOON_API_KEY=db_sk_your_key

import os
from driftballoon import DriftBalloon

db = DriftBalloon(api_key=os.environ["DRIFTBALLOON_API_KEY"])

For self-hosted deployments, set the base URL:

db = DriftBalloon(api_key="db_sk_xxx", base_url="https://driftballoon.your-company.com")

API Reference

DriftBalloon(api_key, base_url=None, sync_interval=30.0, auto_start=True)

Initialize the client. Can be used as a context manager.

log(name, response, prompt=None, model=None) -> LogTask

Log an LLM response. Call .submit() (async, fire-and-forget) or .invoke() (synchronous).

get_active_prompt(name) -> "a" | "b" | None

Get the active prompt version from cached config.

get_config(name) -> PromptConfig | None

Get the full prompt configuration.

get_baseline_status(name) -> (status, count)

Check if the baseline is ready ("learning" or "ready") and how many samples have been collected.

Documentation

Full docs at docs.driftballoon.com.

Local Development

# Setup
make install

# Unit tests (no backend required)
make test

# Integration tests (requires running API)
make test-integration

# Quickstart smoke test
DRIFTBALLOON_API_KEY=db_sk_xxx python examples/quickstart.py

# Cross-venv testing from repo root
make sdk-test-local KEY=db_sk_xxx

License

MIT