Voice AI is scaling infrastructure before vertical workflows
The open market is rapidly expanding recognition, synthesis, and real-time models, but it has barely moved into specialized, regulated workflows. Verticalization remains a hypothesis rather than a validated mass shift.
Verticalization has not happened yet
Specialized workflows represented 8% of the broad GitHub cohort in Q1 and 7.4% in Q2. Universal assistants and infrastructure remained above 92%. Explicit compliance features appeared in only 1% of Q1 projects and were absent from the Q2 cohort.
The universal layer remains dominant
Shares use only the broad voice AI, voice agent, speech agent, and voice assistant cohort.
16 → 12
185 → 151
2 → 0
new projects
Healthcare is the only growing vertical pocket
Categories overlap; small absolute counts require cautious interpretation.
Healthcare rose from two to five projects, but none explicitly stated HIPAA, consent management, audit trails, or another compliance layer. This is early supply, not a mature regulated category.
Open-source infrastructure remains the core market
A layer must be explicitly present in a GitHub name, description, or topic.
Speech recognition held at 45–46 projects, while speech-to-speech rose from 10 to 13. The decline in TTS and telephony cannot be separated from age bias because Q2 repositories had less time to reach three stars.
Hugging Face confirms infrastructure scale, not temporal growth
API cohorts are capped at the latest 1,000 models per filter, so this is a current snapshot of H1 models rather than a comparable Q1/Q2 time series.
624 with downloads
unique accounts
84 models
25 models
* Publishers are summed across quarterly cohorts and may repeat between Q1 and Q2. The full corpus contains 500 unique publishers. Vertical matches (29) are multi-label and do not represent the same number of independent products.
What true verticalization would require
Workflow must become more important than voice
The category will form when products are described through intake, claims, collections, or scheduling rather than STT, TTS, and latency.
Compliance must become a product feature
HIPAA, recording consent, auditability, data retention, and PII redaction must become architectural features rather than external promises.
Healthcare is the earliest candidate
It is the only vertical that grew in the GitHub cohort, but current supply still lacks the required regulatory layer.
Signals tested by this research
A direct link between AILANTA's early observations and the study's measured findings.
A public launch is defined as a new GitHub repository or Hugging Face model publication from January through June 2026. GitHub uses broad and vertical queries capped at two top-by-stars pages per query and month. Hugging Face is classified from metadata and README text. Product Hunt is excluded.
GitHub has an age bias from stars. Hugging Face top-by-recency cohorts do not provide a comparable time series; downloads are current snapshots. Workflow and compliance require explicit contextual evidence and are lower-bound estimates. Verticals overlap.