Monitor a Pipecat agent

Pipecat monitoring sends each production conversation from your bot through the Egma SDK. You do not need a test suite or a simulation connection.

Set up with your coding agent

Open Agents → Connect an agent.
Choose Monitor production, then Pipecat.
Copy the prompt, open your bot’s repository in a coding agent, and paste it.

The coding agent asks where your bot runs in production, adds the SDK line and the settings below, and asks you before it changes a secret set or deploys. Opening the prompt alone does not turn on monitoring. To do the same by hand, follow the steps below.

Set the environment

Create a project API key for production from your initialized agent repository:

egma project api-key create --name "Front desk production"

Copy the key when it is shown; it is shown once. Your bot needs three settings wherever it runs in production. On Pipecat Cloud, add them to the agent’s secret set:

Variable	Value
`EGMA_URL`	`https://app.egma.ai`, or the public URL of your self-hosted Egma instance.
`EGMA_API_KEY`	The project API key you just created.
`EGMA_AGENT_NAME`	Your agent’s name in Egma, as `egma/config.yaml` lists it.

A Pipecat bot has no name of its own, so the SDK sends EGMA_AGENT_NAME with each production conversation, and Monitoring shows it in the Agent column, as it shows a LiveKit worker’s agent_name. Without it, conversations still arrive and are graded, but show no agent name. The bot must be able to reach EGMA_URL.

Add the monitoring line

Install the Pipecat Python SDK:

uv add "egma[pipecat]"

Add await monitor(worker, runner_args) after you create the PipelineWorker and before runner.add_workers(worker). If the bot also runs simulations, keep simulation first:

from egma.pipecat import monitor, simulation


async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
    ...
    worker = PipelineWorker(pipeline, params=PipelineParams(...))

    await simulation(worker, runner_args)
    await monitor(worker, runner_args)

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

Pass runner_args from bot() into the function that builds the worker, as shown in Prepare the bot. Redeploy the bot. The line records each conversation’s turns, tool calls with their arguments and results, and timings, and sends them to Egma. A simulation that simulation reported does not also appear as a production conversation; the SDK guide explains how monitor tells them apart. monitor never stops your bot. If EGMA_URL or EGMA_API_KEY is missing or invalid, it logs a warning and sends nothing.

Verify a conversation

Complete one normal conversation with the updated bot, then open Traces in the project that owns the API key. Open the new trace and check its transcript and timing. There is no Pipecat monitoring switch to turn on in Egma. The first received trace confirms that your bot can send conversations to Egma. If no conversation appears, check the bot’s logs, the project key, and network access to EGMA_URL. To stop sending production conversations, remove the monitor line and redeploy. Existing conversations stay in Egma. Keep simulation if the bot still runs Egma tests.

Review the evidence

Review production conversations to inspect transcripts, metrics, and grades. To grade them, configure graders with production scope.

Get started

Core Philosophies

Platform

Integrations

Support

Monitor a Pipecat agent

Set up with your coding agent

Set the environment

Add the monitoring line

Verify a conversation

Review the evidence

​Set up with your coding agent

​Set the environment

​Add the monitoring line

​Verify a conversation

​Review the evidence

Set up with your coding agent

Set the environment

Add the monitoring line

Verify a conversation

Review the evidence