- aioesphomeapi
- kokoro-onnx
- spaCy (+ thinc, srsly, wasabi, preshed, murmurhash, cymem, catalogue, blis, en_core_web_sm)
- onnxruntime
- onvif-zeep-async
- zeep
- pydantic / pydantic-core
- mlx / mlx-lm / mlx-whisper
- faster-whisper
- tiktoken
- typer
- rich
- attrs
- anyio
- jsonschema
- urllib3
- PyYAML
- go2rtc
- llama.cpp / llama-server
- express
- ws
- serialport
- @noble/curves
- @noble/hashes
- @stoprocent/noble
- @stoprocent/bleno
- usb
- uuid
- jszip (used under MIT)
- …and other transitive dependencies
Open source licenses.
Ambryon is built on the work of the open source community. Below are the third-party components we distribute with the software, grouped by license, with a link to each license’s full text. A complete, versioned inventory — including every transitive dependency — ships with the product.
- matter.js (@project-chip/matter.js, @matter/*)
- matter-server (@matter-server/*)
- aiohttp
- huggingface_hub
- transformers
- tokenizers
- safetensors
- sentencepiece
- fastembed
- regex
- requests
- pylutron-caseta
- astral
- cryptography (Apache-2.0 OR BSD-3-Clause)
- flatbuffers
- @material/web
- @mdi/js
- numpy
- scipy
- torch
- websockets
- protobuf
- Jinja2
- MarkupSafe
- click
- sympy
- networkx
- lxml
- numba
- llvmlite
- Pygments
- lit / lit-html / lit-element
- shellingham
- yaml (JS)
- semver
- typing_extensions
- aiohappyeyeballs
- tslib
File-level copyleft. orjson and tqdm are additionally dual-licensed under permissive terms (Apache-2.0 / MIT); certifi is MPL-2.0 only, which requires sharing modifications solely to certifi's own source files.
- certifi
- orjson
- tqdm
Used as dynamically-loaded, unmodified libraries. The corresponding license texts are included and the upstream source is available at each project. Users may substitute a modified version of the library.
- zeroconf (LGPL-2.1-or-later)
- num2words (LGPL-2.1-or-later)
- WSDiscovery (LGPLv3-or-later)
- pyaarlo (LGPLv3-or-later)
Used by the Kokoro text-to-speech grapheme-to-phoneme path. Ambryon offers the complete corresponding source code for these components under GPLv3-compatible terms on request — contact support@ambryon.ai.
- phonemizer-fork
- eSpeak NG (bundled via espeakng-loader)
On-device models.
The speech and language models that run locally on your Ambryon server. All are distributed under permissive open source licenses.
| Model | Source | License |
|---|---|---|
| Kokoro-82M (text-to-speech) | hexgrad/Kokoro-82M | Apache-2.0 |
| Whisper Large v3 Turbo (speech-to-text) | OpenAI, via mlx-community / faster-whisper | MIT |
| Gemma 4 (12B-it / E4B-it, language model) | Apache-2.0 | |
| YOLOX-Nano (camera person detection) | Megvii-BaseDetection/YOLOX | Apache-2.0 |
To request the corresponding source code for any GPL/LGPL component listed above, contact support@ambryon.ai.