ThunderPhone profile, products and governance

The product is a platform for inbound and outbound AI phone agents with an API, an MCP server and plugins, campaigns of up to 5,000 contacts, live supervision and a white-label option for agencies. Its launch post of 9 November 2026 introduces ThunderPhone v3, featuring Zeus, which ThunderPhone presents as a fully agentic system for voice AI: an AI agent that builds, tests and tunes voice agents, and that the post compares to Claude Code, the coding tool from Anthropic, but for phone calls. The same post says ThunderPhone worked closely with Claude Code and Codex, the coding agent from OpenAI, while building it. With v3 the earlier Spark, Bolt and Storm stacks were retired in favour of one architecture.

That architecture is the company's main claim. Instead of the usual transcription, language model and text-to-speech pipeline, ThunderPhone says it runs several transcription models at once, a faster and a slower language model chosen turn by turn, and speech-to-speech models fused in for natural delivery, in 62 languages. Platform voices are powered by Inworld, with Cartesia and ElevenLabs voices as premium options. Zeus can import existing agents from Vapi, Retell AI, Bland AI and ElevenLabs in one click, according to the post.

On trust, its trust centre states compliance with GDPR, HIPAA and UK GDPR, lists 106 documented controls and 27 subprocessors, and makes 40 policy documents available on request. The homepage adds that a business associate agreement and a data processing agreement come at no extra charge. These are the company's own statements, not audit reports.

On performance, ThunderPhone claims a 99.4% score on Big Bench Audio, which it calls the second-highest published score and the highest of any model fast enough for live calls, and says it handles thousands of real customer calls. It says public versions of its own evals are coming. Funding, investors, headcount and valuation are not stated on its site, so this page leaves them blank.

Where it is weaker: this is a young company whose flagship agent launched in November 2026, and the only public benchmark it cites is one it describes itself. A buyer who wants a long public track record, or who builds voice agents by hand, can also look at the ElevenLabs voice platform. Hokai's guide to AI receptionists covers the most common use case, and other early-stage companies sit in the startup directory.