Test and monitor voice agents in development and production

Egma is an open-source platform for testing voice agents before shipping them and monitoring their behavior once they're in production. For testing, it builds regression suites out of simulated voice and text conversations with mocked tool responses, kept in the agent's own repository and run via the Egma CLI or a coding agent; caller personas and the models used to drive them are customizable, and grading instructions with pass thresholds define what "correct" behavior means for a given test. For production, it lets you review real conversations, tool calls, and metrics, and use graders to check agent behavior and investigate failures. Egma supports LiveKit (JS/TS) and Retell voice-agent platforms today, with other platforms addable on request via Discord. On both Egma Cloud and a self-hosted instance, you bring your own model-provider API keys and pay those providers directly - Egma adds no markup on model usage.

Getting started is a single prompt pasted into a coding agent inside the voice agent's own repository:

Set up Egma in this repo. Install its skills with
npx skills add egma-ai/egma, then use them to connect
my voice agent and run my first voice simulation.

For Egma Cloud, create an account first and use the same prompt; for self-hosting, clone the repo, install the CLI, configure .env per the self-hosting guide, and run egma self-host up, then open localhost:3101 to create an account.

Roadmap: Egma plans to add automatic fixes - using a production failure to propose a change to the voice agent's own harness, test that change, and open a pull request for review.

Use it to build a regression suite for a voice agent built on LiveKit or Retell, and to monitor and investigate that agent's real production conversations and tool calls. It does not (yet) support platforms beyond LiveKit and Retell without a custom request, and its planned automatic-fix feature - proposing and testing a harness change from a production failure - is not built yet, only roadmapped.

Input is a voice agent's codebase (for test setup) or its live production conversation/tool-call stream (for monitoring), plus grading instructions and pass thresholds. Output is a regression test suite with pass/fail grades, or production monitoring dashboards, metrics, and grader-flagged failures for investigation.

Connects to LiveKit (JS/TS) and Retell voice-agent platforms; installs via npx skills add egma-ai/egma as a coding-agent skill, or via the Egma CLI (npm install --global egma-cli) for self-hosting. Bring-your-own API keys for model providers on both Cloud and self-hosted deployments.

Teams building voice agents on LiveKit or Retell who want a regression test suite kept alongside their code, plus production monitoring and failure investigation. MIT licensed (except the ee enterprise-edition folders).