ValGuard.ai logoValGuard.ai

ValGuard + Retell: validate voice agent LLM turns

Configure Retell to call ValGuard as the OpenAI-compatible LLM endpoint. Latency includes model time plus any block or re-ask buffering.

Last verified:

Retell AI builds voice agents with configurable LLM backends. Aim the LLM base URL at ValGuard for the same policy packs you use in chat, and plan for validation latency in conversational UX.

When you need this

Call-center wrap-up JSON includes wrap_up_code and risk_score. Retell forwards model output to your CRM. ValGuard enforces enums and thresholds before the webhook fires.

Setup

  1. Create a ValGuard agent for the Retell LLM step.
  2. Configure Retell's OpenAI-compatible LLM settings to https://api.valguard.ai/v1 with your key and X-VG-Agent.
  3. Shadow on staging agents; watch would-blocks per call type.
  4. Enforce when false positives fit supervisor review load.

Code

curl -s https://api.valguard.ai/v1/chat/completions \
  -H "Authorization: Bearer $VG_API_KEY" \
  -H "X-VG-Agent: retell-wrap-up" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-4o-mini","messages":[{"role":"user","content":"Summarize call outcome as JSON"}]}'

Honest limits

  • Retell-specific features (voice, barge-in, telephony) are not validated by ValGuard.
  • Block and re-ask buffer completions — callers may hear silence until validation finishes.
  • CRM webhooks that never pass through the LLM need their own checks.

Related

Next step

Quickstart.