We build Diction, an iOS keyboard for voice and AI. This org is where the infrastructure behind it lives: the self-hosted server, the container images, and the models it depends on.
- Diction — the open-source gateway and
self-hosted stack.
docker compose upand you have your own speech-to-text server. - Models the app and the self-hosted stack pull from live at huggingface.co/DictionLabs — built for Diction, credited fully wherever the underlying work started as someone else's.
Container images for the self-hosted stack, published under dictionlabs/*:
| Image | Tags | Purpose |
|---|---|---|
ghcr.io/dictionlabs/gateway |
latest, semver |
Diction gateway: streaming proxy, AES-256-GCM encryption, LLM cleanup |
ghcr.io/dictionlabs/whisper-server |
latest-cpu, latest-cuda |
Whisper speech-to-text server |
ghcr.io/dictionlabs/parakeet |
latest-int8 |
NVIDIA Parakeet, 25 European languages, faster than Whisper on CPU |
All three are also on Docker Hub at docker.io/dictionlabs/*. We control what latest points
at and when it moves.