Skip to content

Operator setup

Use this checklist before wiring production traffic.

Workspaces hold agents, users, provider credentials, tools, knowledge sources, voices, traces, and API keys. Use separate workspaces for production and test environments when you need separation.

Configure the speech, model, and voice providers your agent will use. Hyponema supports bring-your-own provider keys. Production deployments should use KMS-backed key encryption for stored credentials.

Write the system prompt, first message, language, guardrails, sleep windows, and escalation behavior. Keep the first persona version narrow enough to test.

Attach the persona to an agent and choose the voice stack. For voice agents, validate STT, LLM, and TTS providers together because latency and fallback behavior are experienced as one system.

Create a test user and run a short session. Verify:

  • The greeting is appropriate.
  • The agent follows the persona.
  • Memory context appears only when relevant.
  • Tool calls are visible in traces.
  • No-go zones redirect or escalate as expected.

Open observability after every test run. Inspect the turn hierarchy, LLM output, tool calls, guardrail results, errors, and latency.

Start with internal traffic or a small beta cohort. Add tools, telephony, and custom greeting behavior only after the base session path is stable.