Observed arrival · 2026-09-18
openvons: The AI That Doesn't Write
An open-source decision layer that classifies voice, images, and text into fixed choices with confidence scores instead of generating prose.
- For
- Teams routing inputs into controlled operational decisions
- Worth noticing
- The site reports 40 ms for eight bundled questions versus 900 ms for generated responses.
Field notes
The project models inputs as finite choice sets rather than open-ended writing. Its examples include routing inquiry emails, selecting stations by voice, navigating 195 river-camera locations, and issuing camera-control commands with confirmation for risky actions. The Android description says recognition runs on-device after a first model download, with a 39 MB arm64 package and no later network requirement. Benchmark figures are project-reported and include synthetic speech and public image datasets.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue