Voice agent graphs are finite state machines. The chip industry has forty years of methodology for verifying FSMs too large to enumerate — structural coverage, functional covergroups, coverage closure. This post maps that methodology onto voice agent testing and shows what voicetest already supports, what's missing, and why the framing matters.
May 15, 2026Import Retell call transcripts as voicetest runs, replay them against the agent's current configuration, and surface behavioral drift before it ships to production.
May 12, 2026Logic, extract, end, transfer, and conversation nodes — plus global interrupts with go-back semantics. The voicetest IR now models the full structure of platform-specific agent graphs without losing fidelity.
May 8, 2026Import from Retell, VAPI, Bland, or LiveKit. Write one test suite. Run it against any agent format. Voicetest's AgentGraph IR normalizes platform differences.
March 15, 2026How voicetest detects duplicated prompt text across agent nodes and extracts reusable snippets to keep your configs maintainable and your tests consistent.
February 24, 2026How voicetest uses a three-model architecture to simulate conversations and evaluate voice agents with LLM-as-judge scoring.
February 18, 2026Route voicetest LLM calls through your Claude Pro or Max subscription instead of paying per-token API costs. No API keys, no proxy — just the claude CLI.
February 16, 2026Set up automated voice agent testing for your Retell Conversation Flow. Import your agent, write test cases, and run them on every push with GitHub Actions.
February 8, 2026