NovaSynth by Noveum

NovaSynth by Noveum is an AI agent simulation and testing platform designed to help developers evaluate voice and chat agents before releasing them to real users. It generates synthetic users, personas, and scenarios that simulate realistic conversations, allowing teams to identify weaknesses, test changes, and improve the reliability of their AI systems. The product is part of Noveum’s broader AI agent reliability platform, which combines tracing, evaluation, simulation, and diagnostic recommendations.

Unlike testing tools that rely exclusively on predefined scripts or mocked conversations, NovaSynth can interact with configured agents through real telephony connections, real-time audio, and chat interfaces. This makes it useful for testing how an agent responds to interruptions, silence, changes in topic, frustrated callers, different speaking styles, and attempts to bypass safety instructions.

The platform allows teams to define synthetic caller personas with specific goals, moods, backgrounds, languages, accents, and conversational behaviours. For example, a customer-support team could simulate a frustrated customer repeatedly asking about a delayed refund, while a developer could test whether a chatbot correctly handles an attempt to reveal confidential system instructions. These scenarios help expose failures that may not appear during ordinary happy-path testing.

NovaSynth supports voice testing through SIP telephony providers such as Twilio, Plivo, and Vobiz, as well as LiveKit real-time audio. It also supports chat over HTTP and integrations with agent frameworks and development environments, including Pipecat, LangChain, LangGraph, and the OpenAI and Anthropic SDKs. Custom-built agents can also be connected through supported SDKs or integration hooks.

Another important capability is batch matrix testing. Teams can run multiple combinations of personas and scenarios under a shared batch identifier, making it easier to compare results and identify patterns across repeated tests. Its asynchronous queue architecture supports parallel execution, and the platform advertises load-testing capabilities of up to 2,000 concurrent lines when appropriately provisioned.

NovaSynth also supports tool virtualization to make testing safer. Read operations can be replayed, while write operations can be sandboxed so test runs do not unintentionally modify production data. This is particularly useful when evaluating agents that interact with external tools, databases, or business workflows.

After each run, NovaSynth captures conversation traces and evaluates results using voice and audio metrics. These measurements help teams investigate problems involving response quality, turn-taking, latency, and other aspects of conversational behaviour. Within Noveum’s wider platform, these results can support further investigation and validation of fixes.

NovaSynth is primarily designed for AI engineers, software developers, quality-assurance teams, and businesses deploying customer-service chatbots, voice assistants, and multi-agent workflows. It is particularly relevant to organisations that need repeatable testing before release, frequent regression testing after prompt or code changes, and more systematic coverage of unusual conversational situations.

The product is a pre-production testing tool, so it is intended to evaluate agents in test or staging environments rather than automatically placing simulated callers into live production systems. Its value depends on the quality of the configured scenarios, the agent context provided, and the metrics selected for evaluation. Teams should also verify that test environments and sandboxing are correctly configured before running scenarios that interact with external systems.

Pricing

NovaSynth is included in Noveum’s unified subscription and credit system. Its voice and text testing capabilities are available across the platform’s plans, including the Free plan, but actual testing consumes credits.

• Free plan: $0 per month, with 2,500 credits per month.
• Pro: $69 per month, with 10,000 credits per month.
• Growth: $99 per month, with 25,000 credits per month.
• Max: $199 per month, with 50,000 credits per month.
• Scale: $599 per month, with 200,000 credits per month.
• Enterprise: Custom pricing, with unlimited credits and usage capacities subject to the agreement.
• Free trial: A seven-day trial is advertised for the Pro, Growth, and Max plans.

NovaSynth testing uses 25 credits per minute for voice calls and 10 credits per text conversation. Additional credit packs are available as one-time purchases: Boost provides 5,000 credits for $35, Accelerate provides 25,000 credits for $150, and Scale provides 100,000 credits for $500. Purchased add-on credits do not expire, while monthly plan credits reset each billing cycle.

Annual billing offers approximately 17% savings compared with monthly billing. Prices are listed in US dollars and may change, so users should confirm the current rates before subscribing.

Review

NovaSynth is a specialised solution for teams that need to test AI agents under realistic conversational conditions. Its strongest feature is the ability to simulate voice and chat interactions with synthetic users whose behaviours extend beyond straightforward scripted exchanges. This makes it useful for finding problems involving interruptions, unexpected requests, emotional conversations, and attempts to circumvent an agent’s instructions.

Its support for real SIP calls and real-time audio is another advantage. Voice agents can behave differently in real conversations than they do in text transcripts, particularly when latency, silence, interruptions, and turn-taking affect the interaction. Testing with actual audio and telephony connections provides a more realistic evaluation environment than relying exclusively on simulated transcripts.

Batch matrix testing and reusable personas also make NovaSynth suitable for regression testing. Development teams can run consistent combinations of scenarios after modifying prompts, tools, or workflows and compare the resulting behaviour. Tool virtualization adds another useful layer by helping teams test tool-using agents without allowing simulated writes to affect production systems.

The main consideration is that NovaSynth is a technical platform rather than a simple no-code chatbot builder. Teams may need to configure endpoints, provide agent context, establish scenarios, select appropriate metrics, and interpret test results. Its credit-based billing also means that intensive voice testing can consume a significant portion of a monthly allowance. Organisations should estimate their expected test duration and volume before choosing a plan.

Overall, NovaSynth is best suited to engineering and quality-assurance teams building voice agents, chatbots, and multi-step AI workflows. Its realistic simulations, batch testing, trace capture, and integration options make it a useful component of an AI reliability workflow. It is less suited to users seeking a general-purpose chatbot or a simple consumer-facing content-generation tool.