Quickstart
Run your first voice simulation in as little as five minutes with Egma’s skills and your coding agent. Open your voice agent’s repository and give your coding agent this prompt:<your-instance-url>” to the prompt above.
Follow the prompts to install the skills for your coding agent and authorize Egma login. Review the proposed test and mock any tools that could change your data. Open the results link to review the conversation and grading results once they are ready.
For LiveKit, the skill may use a local worker for your first run. Deploy any SDK changes before starting later runs.
Prefer to use the terminal yourself? Get started with the CLI.
Testing voice agents
Run simulated voice and text conversations against your agent. Save tests in your agent’s repository and run them again when you change its prompts, models, or tools.- Build regression suites around the tasks your agent needs to handle.
- Mock selected tool responses to test failures and other situations that are hard to reproduce.
- Review transcripts, voice simulation recordings, and grading results.
Monitoring voice agents
Real conversations can expose behavior you didn’t see in testing. Collect production traces to review transcripts, inspect tool calls, and check available performance metrics. Apply graders to check conversations against your agent’s requirements. Egma supports monitoring for LiveKit agents through its Python and JavaScript SDKs, and for Retell agents through a direct integration. Set up monitoring →Why Egma?
When we were building voice products, we spent hours making test calls and investigating problems in production. Every change meant checking the agent’s behavior again. Even after all that testing, real conversations exposed behavior we hadn’t anticipated. We built Egma to make those checks repeatable and help us understand failures in production. We wanted to ship voice agents with more confidence and keep checking how they behaved once real people were using them.- Open source. Inspect the code, adapt it to your workflow, and self-host the platform.
- Bring your own keys. Use your provider keys in Cloud or a self-hosted deployment and pay providers directly for those model calls. Cloud plan and applicable voice usage charges still apply.
- Choose your models. Select from supported language and speech models to suit your tests.
- Customize personas. Set the simulated caller’s behavior, language, and voice to reflect the people your agent serves.
- Customize graders. Write your own grading instructions and set models, pass thresholds, and which conversations to grade.
- Keep tests with your code. Review test changes alongside agent changes and run them from the CLI or your coding agent.