feat: Add pipecat integration

[Abhijeet Prasad (AbhiPrasad)]

resolves https://linear.app/braintrustdata/issue/SDK-7/braintrust-pipecat-integration

AI Summary

Adds observer-based Pipecat instrumentation that injects a
BraintrustPipecatObserver into PipelineWorker construction via setup_pipecat()
and auto_instrument(). The observer turns real Pipecat pipeline/provider events
into Braintrust task and llm spans, with VCR-backed OpenAI/Pipecat coverage for
both latest and 1.3.0.

The integration also supports setup_pipecat(capture_audio=True) for capturing
Pipecat audio attachments in traces. When enabled, user/input audio and TTS audio
chunks are accumulated from real Pipecat frames, wrapped as WAV/PCM Braintrust
Attachments, and logged with audio metadata like sample rate, channel count,
frame count, and byte size.

Try it in an app:

import braintrust
from pipecat.frames.frames import EndFrame, LLMContextFrame
from pipecat.pipeline.pipeline import Pipeline
from pipecat.pipeline.worker import PipelineParams, PipelineWorker
from pipecat.processors.aggregators.llm_context import LLMContext
from pipecat.services.openai.llm import OpenAILLMService
from pipecat.workers.runner import WorkerRunner

braintrust.init_logger(project="pipecat-demo")
braintrust.auto_instrument()  # installs the Pipecat PipelineWorker observer

llm = OpenAILLMService(
    api_key="...",
    settings=OpenAILLMService.Settings(
        model="gpt-4o-mini",
        temperature=0.0,
        max_completion_tokens=64,
    ),
)
worker = PipelineWorker(
    Pipeline([llm]),
    params=PipelineParams(enable_metrics=True, enable_usage_metrics=True),
)
context = LLMContext(messages=[{"role": "user", "content": "Say hello"}])

@worker.event_handler("on_pipeline_started")
async def on_pipeline_started(_worker, _frame):
    await worker.queue_frames([LLMContextFrame(context), EndFrame()])

runner = WorkerRunner(handle_sigint=False)
await runner.add_workers(worker)
await runner.run()

To enable audio attachments explicitly:

from braintrust.integrations.pipecat import setup_pipecat

setup_pipecat(capture_audio=True)

To validate locally from this repo:

cd py
nox -s "test_pipecat(latest)"
nox -s "test_pipecat(1.3.0)"
OPENAI_API_KEY=sk-test-dummy-api-key-for-vcr-tests \
  nox -s "test_pipecat(latest)" -- -k test_setup_pipecat_traces_real_pipeline_frames

Expected trace shape:

pipecat_pipeline                         task
├── user_speaking                        task  input.audio: Attachment
├── tts_response                         task  output.audio: Attachment
└── pipecat_llm_response                 llm
    ├── input:  [{role: developer}, {role: user, ...}]
    ├── output: [{index: 0, finish_reason: stop, message: {...}}]
    ├── metadata: {provider: openai, model: gpt-4o-mini}
    └── metrics: {prompt_tokens, completion_tokens, tokens, time_to_first_token}