Skip to main content
First connect your Pipecat agent and add the SDK line to your bot.

​
Add a chat connection

A chat simulation tests your bot’s reasoning and tools without speech. Egma starts your bot with the same start request, and the persona sends each turn as text through RTVI. It reads your bot’s replies from RTVI’s text messages. Chat needs no code beyond the SDK line. In a chat simulation, the SDK turns your bot’s speech output off, so the greeting and every answer stay text only. Chat does need RTVI, which PipelineWorker turns on by default. Do not pass enable_rtvi=False, and if you pass your own rtvi_observer_params, keep bot_llm_enabled on: Egma reads the bot’s answers from those messages. A bot with RTVI turned off still passes voice simulations, but its chat simulations fail. Add a chat connection beside your voice connection:
For a self-hosted starter, use --access pipecat-self-hosted with --pipecat-start-url, as in Use a self-hosted starter. egma agent dev creates a chat connection for this machine for you. Use this connection’s ID when you start a chat run. The same suite can run against either connection.

​
Pass data to a test

If your bot reads startup data from runner_args.body, put that data in the test’s ## Env section:
Egma adds this object to the body of the start request. Your bot reads it as it does in production:
  • Use any JSON values, with the exact keys your bot reads.
  • The object can be at most 512 KiB once serialized. Keep a larger value in your own store and pass its ID.
  • The key egma is reserved. Egma adds its own egma key to every start request, and refuses a test that holds one in pipecat_body_params.
  • A test without this field sends a body that holds only Egma’s key.
Do not use the egma key to decide whether a conversation is a simulation: any client can send one. simulation asks Egma to accept the bot’s report before it acts on the key, and monitor stays silent only after that accepted report.