NovaSynth by Noveum

No. NovaSynth is pre-production. Our agent calls the agent you point us at - a test number, a staging SIP, or a new agent you have not shipped yet. Start there, see the value, then decide where else to use it. Live production evaluation is a separate post-production module that does not simulate callers.

For a single-agent architecture, yes - put it in Agent config, along with your PRD, BRD, workflow docs, or anything that explains what your agent should do. Better context means better personas, scenarios, and recommendations. For multi-agent, do not paste a single prompt; integrate our SDK so we take everything from your traces.

Give us context instead. Describe what the agent does - loan approval, debt collection, customer support - and we work from that. You still get simulations, evaluations, and a report. Recommendations will be less precise without visibility into the agent's decision-making.

Three common paths. If your agent has a phone number, add it as a phone SIP connection, pick inbound or outbound, choose the country - that is what most customers use. If you are on a framework, connect LiveKit (or Pipecat and similar) directly. For chat agents, add an HTTP endpoint. For multi-agent architectures, use our SDK.

Yes, but integrate. For a single agent you can paste a prompt and run. For multi-agent, put our SDK in - a pip install and a few lines on LangChain, LangGraph, LiveKit, or Pipecat, plus custom hooks if you built ground up. Without it we cannot tell which sub-agent's prompt to fix, or whether the tool call or the agent was at fault.

Yes - that is where this started. You pick noise type (crying baby, cafe, traffic, people talking) and modulate the level per persona. One customer used to send someone to a construction site to call their agent; now it is a click. Dialect, region, and speech patterns go in the persona description.

Yes. We support up to 2,000 concurrent lines for load testing - enough to see how many calls your agent can hold at once. Tell us before a big run so we can provision lines. For ordinary functional tests, keep concurrency low for cleaner results and fewer simulated minutes.

Instruction adherence, hallucination, drop-off detection, sentiment and CSAT, answer relevancy and completeness, role adherence, task progression, content moderation and harmful advice, plus voice-specific metrics: latency, pronunciation, clarity, volume, pitch, MOS, gibberish detection, and STT efficacy. Most are LLM-judged; some, like latency, are computed. We publish 30+ voice AI metrics and 100+ across the platform.

Every call is evaluated on everything, but you choose which metrics affect call success. Scores run 1 to 10 with a default threshold of 7 - below that needs attention. If any of your chosen success metrics drops below threshold, the call is marked a failure. Thresholds are configurable per team and per metric (including latency).

You get a NovaPilot report: categories of failure ranked by criticality, how many calls each one affected, root cause, and the specific change - remove this line from your prompt, add this instead, or fix pipeline, architecture, or model. Download it as markdown, hand it to Cursor or Claude Code, and improve the agent. Then re-run the same batch to see scores move.