Callstack

Callstack is a visual regression and testing platform for voice AI agents built on AssemblyAI. It enables developers to create, run, and compare conversational test cases against different versions of their voice agents before changes reach production. Developers define test cases using real or scripted audio conversations, expected agent behavior, assertions, and configuration. Each test streams audio through AssemblyAI’s real-time speech-to-text infrastructure while Callstack captures the complete execution: transcripts, speaker turns, timestamps, tool calls, agent responses, latency, and other runtime events. The platform turns these executions into visual conversation traces, making it easy to inspect exactly how an agent behaved. Developers can run the same test against two agent versions and view them side-by-side to identify regressions such as incorrect transcription, missed entities, failed tool calls, unexpected responses, increased latency, or premature interruptions. Callstack also provides configurable assertions for validating intent detection, entity extraction, tool usage, response quality, latency thresholds, and expected conversational behavior. Failed assertions are surfaced alongside the relevant section of the conversation, making debugging significantly faster. The pipeline integrates with CI/CD so voice-agent tests can automatically run whenever a prompt, model, tool, orchestration layer, or configuration changes. Teams can establish a baseline, execute a regression suite, compare results, and block deployments when critical conversational behavior degrades. Callstack brings the workflow of Postman and visual regression testing to voice AI: define conversations, execute them, inspect what happened, compare versions, and catch regressions before users do.