Open-Source & Model Attributions

Effective date: 2026-05-30 Last updated: 2026-06-06

dicted is built and delivered by M37 Tech Pty Ltd (ABN 88 696 420 363). Parts of the Service incorporate, build on, or run alongside third-party open-source software and open-weight machine-learning models. We are grateful to the people and organisations who publish them.

The trained weights, fine-tuning, architecture choices, prompts, and integration work that are ours remain our intellectual property (see Terms of Service §7). The components listed below are not ours — they remain the property of their respective authors and are used under the licences shown. Nothing on this page overrides the rights granted to you, or to us, under those upstream licences. Where a component is shipped inside one of our applications, the applicable upstream licence travels with it.

This page is published so that our use of these components is transparent and so that the "proprietary models" language in our Terms is accurate rather than over-claiming.

Machine-learning models

Component	Where it is used	Upstream owner	Licence
Gemma (self-hosted, server-side LLM)	Meeting summaries, minutes, action items, smart actions, Smart Polish, Voice Edit — runs on M37-operated infrastructure; transcript text never leaves our network	Google	Gemma Terms of Use (open-weight, with the Gemma Prohibited Use Policy)
FLAN-T5 (components in the text pipeline)	Sequence-to-sequence text transformation used in parts of the smart-action text pipeline	Google	Apache License 2.0
NVIDIA Parakeet (Parakeet-TDT speech recognition)	Primary speech-to-text for dictation and meeting transcription on our self-hosted pipeline	NVIDIA	CC-BY-4.0 (model weights); NVIDIA NeMo toolkit under Apache License 2.0
pyannote.audio (speaker diarization)	Speaker segmentation / "who spoke when" in meeting transcripts	pyannote / Hervé Bredin and contributors	MIT License (toolkit); pretrained models under their respective model cards

Cloud fallback providers (used only when a primary path is at capacity or unavailable) and infrastructure providers are listed separately in our Subprocessors page and Privacy Policy §6.

How to reach us

Questions about attributions or licensing of components in dicted: [email protected].

We keep this page current as the model components in the Service change. If you believe a component we use is missing from this list or attributed incorrectly, please let us know at the address above.