Stage 3 · v2Coming soon
A local AI layer you choose the brain for.
A swappable on-device model layer — pick the lightweight local model that fits your RAM and GPU, starting with caption punctuation and cleanup.
Gemma 4 · 2BInstalled · 1.6 GB
Gemma 4 · 9B8 GB RAM · Download
Whisper large-v3Downloading · 62%
Feature roadmap — everything planned, honestly labeled
What Hermite AI will do
Model picker & manager
Supported local models listed with disk and RAM requirements shown before download, plus version management. Swap models without touching the rest of the app.
Caption punctuation & cleanup
The first concrete AI use case: a bounded, low-risk pass that directly improves transcription output.
Optional cloud pass (opt-in)
A clearly labeled, off-by-default higher-quality cloud option for users who want it. Never a silent fallback — your data stays local unless you explicitly opt in.
Honest about hardware. Local models cost RAM and disk, and running them next to Resolve is a real budget. The model manager shows requirements before you download, defaults to the smallest viable model, and never installs anything silently. If you opt into the cloud pass, it's a visible switch — never a fallback.
Watch it come together
Hermite is built in the open — the roadmap and changelog update as the code does.