Egma
GitHub Try Egma Cloud

Ship reliable voice agents

Open-source platform for voice agent testing and production monitoring that finds unknown issues proactively.

Try Egma Cloud → View on GitHub

Catch issues before real users do

Define tests as markdownn alongside your agent code. Egma runs realistic text and voice conversations against your agent and evaluates what happened.

See how testing works →
LOCAL REPOSITORY main
EXPLORER VOICE-AGENT
voice-agent
egma
config.yaml
tests
appointment-booking
suite.yaml
book-consultation.md
no-available-slots.md
book-consultation.md
  1. ---
  2. format: 5
  3. name: book-consultation
  4. description: A caller books an available consultation.
  5. personas:
  6. - name: Everyday caller
  7. ---
  8.  
  9. ## Scenario
  10.  
  11. Book an afternoon consultation at Harbor Clinic.
  12. Choose 15:00 when the receptionist offers it.
  13.  
  14. ## Expected behaviors
  15.  
  16. 1. The agent calls check_availability.
  17. 2. The agent offers only available times.
  18. 3. The agent confirms the selected time.
  19.  
  20. ## Mock tools
  21.  
  22. ### check_availability
  23. { "answer": { "slots": ["15:00", "16:30"] } }
  24.  
  25. ## Env
  26. { "job_dispatch_metadata": { "clinic_name": "Harbor Clinic" } }

Know exactly what went wrong

Define in natural language how your agent should behave in each scenario. Egma runs the simulation and grades every conversation against expected behaviors.

See how evaluation works →
EXPECTED BEHAVIORS COMPLETE

Two of three behaviors passed

0.67THRESHOLD 1.00
01 Confirms patient details before offering slotsTranscript · turns 01—03 PASS
02 Only shows slots from existing providerTool result · turn 05 PASS
03 Confirms only after book_appointment tool succeedsTool result + transcript · turns 07—08 FAIL

Find the issues you didn't know to look for

Egma proactively analyzes production conversations to surface unexpected behavior, patterns, and failures.

See how monitoring works →
TRANSCRIPT00:08—00:16
00:08CALLER

“Can you move my appointment to tomorrow after six?”

00:13TOOL

book_appointment

REJECTED / OUTSIDE_BUSINESS_HOURS
00:16AGENT

“Absolutely — I’ve moved it to 6:30 PM.”

Stop paying inference premiums

Run simulations with your own own STT, LLM, and TTS keys or use Egma inference at 0% markup. Stop paying heavy premium to run tests!

Bring your own keys →

Tests live with your code

Author tests alongside code in your repository. Review and version them in git. Ensure your coding agents has full context of how your agent should behave.

Run your first suite →

Open source

You should be able to inspect, modify, and improve the system that you rely on for your production agent’s testing and monitoring.

—

Self-host for enterprise control

Voice agents are already handling sensitive data across industries. Own your production data in your VPC. No long vendor review cycles and lengthy security questionnaires.

Read the self-hosting guide →

Voice agents you can trust

Try Egma Cloud → Read the docs