Why this exists.
I built voice agents for two years before SpeechStack. Every project started the same way: open seven browser tabs, read seven sets of vendor docs, pick a stack, hope it works.
The stacks were the bottleneck. Not the code, not the model. The stack. Which STT pairs well with which TTS at which latency? What does it cost to run a Vapi agent end-to-end with Deepgram and Cartesia? Why does this Retell config feel three turns slower than the LiveKit one? Nobody answered these questions in one place. The vendor docs each made their own product look like the right pick.
SpeechStack is the directory I wanted to read.
Cursor Directory mapped the AI coding stack into one neutral surface. The voice AI category needs the same thing, and it doesn't exist yet. So: this site. Every template shows its stack, prompt, and config. Every template publishes what it costs to run. Every vendor across the five layers gets the same card shape on /tools. Nobody pays for a better position. There isn't one.
The bar. Every template has to satisfy four criteria: schema-able artifact, forkable or copyable, specific named outcome, multi-component stack. If a submission fails any of them, it doesn't ship. Not as a draft, not as a “coming soon.” The bar is the product. Lowering it would turn this back into the thing we're trying to replace.
We don't write editorial. We don't rank vendors. We don't run “best of” lists. The closest we get is /compare/[a-vs-b] programmatic pages, where the comparison is the data, not the prose.
Sponsorship pays for this site (eventually). Sponsors get logo placement on /tools and footer presence. They don't get editorial influence, ranking changes, or template-selection bias. If that trade-off doesn't work for a vendor, the trade-off isn't for them.
SpeechStack is built on the side, by one person, for builders shipping voice AI for real. I made the call to keep it neutral, structured, and honest. If it gets useful, that's the bar holding.
Founder
Zach Capshaw is the founder of SpeechStack. He co-founded Flashquotes, where he's CEO and CMO, and spent the prior decade in software product strategy. He started SpeechStack after building voice AI agents on the side and finding no neutral place to read other people's stacks. He maintains the directory, reviews every submission, and tests verified templates personally.
How templates get added
Two paths: open a PR on GitHub, or fill the form. Both land in the same queue. We review every submission against the four-criteria bar. Approved templates ship with verified:false until Zach test-runs them. The flag flips to true once he's placed a call or run the workflow himself.