Voice AI Day is baaaaack! 🎙️ 10 cities, 10 Dec
    Register here →
    Voice AI / Space
    Submit
    BackSupported byLiveKit
    NovaSynth

    NovaSynth

    Visit Website
    Eval
    Automated Testing
    Security
    Quality Assessment

    Test voice and chat agents with synthetic users before release.

    Founded 2023San Francisco, California, United StatesLinkedIn
    NovaSynth banner

    About NovaSynth

    NovaSynth: Synthetic testing for voice and chat agents

    NovaSynth tests pre-production voice and chat agents with synthetic users over real phone calls, real-time audio, and chat. Teams configure personas and scenarios to surface failures, evaluate complete conversations, and validate candidate fixes before release.

    Key Features

    • Real voice connections: Run SIP calls through Twilio, Plivo, or Vobiz and LiveKit real-time audio, with real telephony, audio, and latency rather than mocked transcripts.
    • Agent integrations: Connect Pipecat, LangChain/LangGraph, OpenAI and Anthropic SDKs, chat over HTTP, or custom agents through an SDK or light hooks around LLM calls.
    • Generated personas and scenarios: Generate synthetic users with goals, moods, accents, and behaviors, using agent context such as prompts, requirements, workflows, and documents.
    • Conversation disruptions: Test interruptions, barge-in, silence, topic switches, and users who switch languages mid-sentence.
    • Batch matrix runs: Run combinations of personas and scenarios under a shared batch ID to keep coverage grouped and comparable.
    • Parallel execution: An asynchronous queue schedules and parallelizes hundreds of scenarios per run without blocking your environment.
    • Load testing: Raise concurrency to test up to 2,000 simultaneous lines, with advance notice required for large load runs so lines can be provisioned.
    • Tool virtualization: Replay reads and sandbox writes to exercise changed decisions end-to-end without sending writes to production systems.
    • Trace capture and scoring: Capture every synthetic call as a full trace and score it, including the audio, with more than 30 voice AI metrics.

    Use Cases

    • Pre-release testing: Point NovaSynth at a test number, staging SIP endpoint, or unshipped agent to uncover failures before customers encounter them.
    • Caller recovery: Rehearse frustrated repeat callers and rapid interrupters to test de-escalation, turn-taking, and recovery under pressure.
    • Safety testing: Use prompt-injection scenarios that request system prompts and tool definitions to test safety boundaries and refusal behavior.
    • Fix validation: Test team-reviewed candidate changes with configured voice and chat agents, personas, and scenarios.

    Getting Started

    NovaSynth combines synthetic callers, real voice and chat connections, and scored traces to test agent behavior before release.