Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.tryelixir.ai/llms.txt

Use this file to discover all available pages before exploring further.

1

Install Elixir

pip install elixir-ai
2

Set up your project

You’ll need an Elixir API key to use the SDK and begin logging data.
  1. Create an Elixir account.
  2. Create a new project.
  3. Navigate to project settings and issue a new API key. title
3

Set up your environment

.env
...
ELIXIR_API_KEY=<your-api-key>
4

Trace your agent conversation

  1. Call Elixir.init() to instrument calls to your LLM services.
(This uses an OpenTelemetry-compatible tracing standard. Asynchronous instrumentation ensures this will not add latency to your service.)
# import Elixir
from elixir import Elixir

# initialize globally
Elixir.init()
  1. Call Elixir.track_conversation(call_id: str) to track the current conversation and group LLM traces within. You’ll need to provide a call id.
Examples:
pipecat.py
from pipecat.frames.frames import Frame
from pipecat.processors.frame_processor import FrameDirection
from pipecat.services.openai import OpenAILLMService as BaseOpenAILLMService
from elixir import Elixir


class OpenAILLMService(BaseOpenAILLMService):
    def __init__(self, session_id: str, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.session_id = session_id

    def process_frame(self, frame: Frame, direction: FrameDirection):
        # Pass conversation id here
        Elixir.track_conversation(self.session_id)
        return super().process_frame(frame, direction)
custom_llm.py
@custom_llm.route("/chat/completions", methods=["POST"])
def custom_llm_openai_sse_handler():
  data = request.get_json()
  Elixir.track_conversation(data["call"]["id"])
5

Send audio to Elixir

After the call ends, use Elixir.upload_audio(conversation_id: str, audio_url: str) to send call recording to Elixir. This can be used in one of two ways:
  1. Send a publicly accessible call recording URL.
  2. Upload the file directly (if recording link is not public).
Examples:
bot_runner.py
@app.post("/twilio_recording", response_class=PlainTextResponse)
async def twilio_recording(request: Request):
    print("POST /twilio_recording")

    data = {}
    try:
        # shouldnt have received json, twilio sends form data
        form_data = await request.form()
        data = dict(form_data)
    except Exception:
        pass

    callId = data.get("CallSid")
    recordingUrl = data.get("RecordingUrl")

    if not callId or not recordingUrl:
        raise HTTPException(
            status_code=500, detail="Missing 'CallSid' or 'RecordingUrl' in request"
        )

    # Download the recording from Twilio
    async with aiohttp.ClientSession() as session:
        async with session.get(
            f"{recordingUrl}.mp3?RequestedChannels=2",
            auth=aiohttp.BasicAuth(TWILIO_ACCOUNT_SID, TWILIO_AUTH_TOKEN),
        ) as response:
            if response.status == 200:
                recording_content = await response.read()
                content_type = response.headers.get("Content-Type")

                print(
                    f"Recording content: {content_type}, {len(recording_content)} bytes"
                )

                await Elixir.upload_audio(
                    conversation_id=callId,
                    audio_buffer=recording_content,
                    audio_content_type=content_type,
                )
            else:
                raise HTTPException(
                    status_code=500, detail="Failed to download recording from Twilio"
                )

​
Additional Resources

SDK Guide

Learn how to use the Elixir SDK