Third party notices

The AI NPC Cyberpunk Mod License (LICENSE) covers only this project's own source code, documentation and configuration. It does not replace the licences below, and it grants nothing in Cyberpunk 2077 or its assets. Each release package also carries these notices as THIRD-PARTY-NOTICES.txt (generated by packaging/build-release.mjs with the exact versions shipped) next to the components' own licence files.

This file states what each component's published licence says. It is not legal advice.

Bundled in the Runtime package#

vgmstream (V voice mods)#

The Runtime carries vgmstream-cli.exe and ten libraries it loads in app/vgmstream/: the unmodified files of the official vgmstream Windows release r2055, pinned by size and SHA-256 in packaging/release.config.json (toolchain.vgmstream). The mod starts it only to decode the voice-over of a V voice mod that the player installed, one line at a time, on the player's own PC, to build V's voice reference; the process exits when the build is done and nothing is sent anywhere. The audio it decodes belongs to the player's game and voice mod, not to this package. vgmstream is ISC-licensed; the FFmpeg and mpg123 libraries are LGPL-2.1-or-later and are used as separate DLLs, unmodified and dynamically linked, so a player can replace them with another build. Their licences and the source locations are in app/licenses/vgmstream/ (COPYING, NOTICE-bundled-libraries.txt); the sources and build scripts of every library are in the vgmstream repository at the tag r2055. The G.719 library is the one component whose licence text the release does not give.

Lucide icons (character editor)#

The character editor page (bridge/character-editor-page.mjs, the web page the Character editor row opens on the player's own PC) draws its icons from the Lucide icon set (https://lucide.dev). Only the shapes the page uses are copied into the page as inline SVG markup, so the page fetches nothing and needs no icon font, library or network. Lucide is ISC-licensed, and the parts of it that come from Feather are MIT-licensed. Both licence texts follow, as Lucide publishes them.

ISC License

Copyright (c) for portions of Lucide are held by Cole Bemis 2013-2022 as part of Feather (MIT). All other copyright (c) for Lucide are held by Lucide Contributors 2022.

Permission to use, copy, modify, and/or distribute this software for any purpose with or without fee is hereby granted, provided that the above copyright notice and this
permission notice appear in all copies.

THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO
EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.

The MIT License (MIT) (for portions derived from Feather)

Copyright (c) 2013-2022 Cole Bemis

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software
without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit
persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR
PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Pocket voice model (Kyutai)#

The Pocket voice model in app/models/pocket/ is built from Kyutai's published Pocket TTS export (kyutai/pocket-tts). What the sources say:

• The code (github.com/kyutai-labs/pocket-tts) is MIT-licensed, including at the pinned revision.
• The Hugging Face model card for kyutai/pocket-tts declares the weights CC-BY-4.0. The repository is gated: access requires accepting the card's prohibited-use terms, which list voice impersonation or cloning without explicit and lawful consent among the prohibited uses.
• Per receipt.json and decoder-receipt.json, the shipped export takes its Mimi codec, inference.yaml and tokenizer.json from the template export, replaces the flow language-model tensors with weights trained for this mod, and replaces 44 Mimi decoder tensors with a retrained decoder. CC-BY-4.0 asks for attribution to Kyutai, a link to the licence (https://creativecommons.org/licenses/by/4.0/) and an indication that changes were made. This section is that notice: the Pocket voice model is adapted from Kyutai's Pocket TTS.

Voice reference clips and game archives#

The Runtime package carries reference voice clips (voices/) and three game archives built from Cyberpunk 2077 files with WolvenKit (archive/pc/mod/). The game, its audio and its voice performances belong to CD PROJEKT RED and the performers. The AI NPC Cyberpunk Mod License does not grant rights in them.

Child voice reference clips#

The Runtime carries four short reference clips, voices/child-voice/child-boy.wav, child-girl.wav, child-girl-2.wav and child-girl-3.wav (9 to 10 seconds each, 24 kHz mono), with a manifest (child-voice.json). They give the voice of a child NPC (a boy or a girl among the crowd) to the voices that clone from a clip. A girl gets one of the three girl clips, chosen from her stable identity, so the same girl always keeps the same voice. Each clip is the same two spoken sentences about a frog, generated with Cartesia text to speech (https://cartesia.ai, model sonic-3) from stock voices of its library: Cathy - Coworker (e8e5fffb-252c-436d-b842-8879b84445b6) for the boy, Daisy - Reading Girl (32b3f3c5-7171-46aa-abe7-b598964aa793), Avery - Gaming Girl (cccc21e8-5bcf-4ff0-bc7f-be4e40afc544) and Katie - Friendly Fixer (f786b574-daa5-4673-aa0c-cbe3e8534c02) for the girls.

• Source: generated with Cartesia, no recording of a real person is in the clips. Changes made: leading and trailing silence cut, level set to 80 percent of full scale, 10 ms fades; the audio was produced at 24 kHz, 16-bit mono and is otherwise unchanged.
• Terms: the generated audio comes from the Cartesia account of the mod's owner; Cartesia's terms of service apply to it.

Trained models#

The Pocket voice and lip-sync models were trained for this mod. The existing notice reads: "trained for this mod; free for personal use with the mod." They are model weights, not source code, so the AI NPC Cyberpunk Mod License does not by itself set their terms; this notice and the Kyutai terms above do.

Audio helper#

app/audio/CyberpunkConvaiAudio.exe is built from native/audio in the release build (it comes with the Scripts, next to the bridge). It uses only the Windows audio APIs (WASAPI, the Voice Capture DSP) and the C++ runtime linked statically (MSVC); it contains no third-party code and needs no licence text of its own beyond this project's.

LiveKit native binding (WebRTC, FFmpeg, OpenH264)#

app/node_modules/@livekit/rtc-ffi-bindings-win32-x64-msvc/rtc-node.win32-x64-msvc.node is LiveKit's own build, shipped unmodified as npm publishes it (version 0.13.1). It links Google's WebRTC build webrtc-89d790b statically, and with it the libraries WebRTC is built with. LiveKit built it with rtc_use_h264 = true and ffmpeg_branding = "Chrome", so it holds FFmpeg's decoders as Chromium builds them (H.264, AAC, MP3, Vorbis, FLAC among them) and an OpenH264 encoder compiled from source. The mod itself sends Opus audio and its picture as a video track (codec chosen by LiveKit), it does not ask for H.264.

• FFmpeg is under the GNU Lesser General Public License 2.1 or later. Its complete source is Chromium's FFmpeg at the revision WebRTC webrtc-89d790b pins (https://chromium.googlesource.com/chromium/third_party/ffmpeg), and LiveKit's build scripts and patches are public (https://github.com/livekit/rust-sdks, webrtc-sys/libwebrtc), so the binding can be rebuilt with a changed FFmpeg.
• OpenH264 is BSD-2-Clause, Copyright (c) 2013 Cisco Systems. Cisco's patent licence for H.264 covers only Cisco's own OpenH264 binaries, not one compiled into another program, and the same holds for the H.264 and AAC decoders of FFmpeg: no patent licence comes with this binding.
• Every licence text of the build is in app/licenses/livekit/WebRTC-third-party-licenses.md (LiveKit's LICENSE.md of that build, made by WebRTC's generate_licenses.py).

Native core plugin#

red4ext/plugins/AiNpcCyberpunkModNative/CyberpunkConvaiNative.dll, built from native/core in the release build (part of the Runtime package). Both components are compiled into the DLL; their licence texts ship next to it (RED4ext.SDK-LICENSE.md, yyjson-LICENSE.txt).

Bundled in the optional OmniVoice pack#

The OmniVoice weights are CC-BY-NC, which excludes commercial use. The pack is optional and separate from the Runtime package.

Bundled in the optional Local AI packs#

Two optional downloads, "Local AI Ultra" and "Local AI". Both unpack into red4ext/plugins/AiNpcCyberpunkMod/app/local-ai/ and each carries the same licence texts and THIRD-PARTY-NOTICES.txt. Every file is the unmodified upstream file, pinned by revision and SHA-256 in packaging/release.config.json. The licences below are what each publisher's model card or repository states; the Gemma 4 card says Apache-2.0 and links Google's Apache-2.0 page, and Google also publishes a prohibited use policy for Gemma (https://ai.google.dev/gemma/prohibited_use_policy).

Apache-2.0 and MIT both allow redistribution, including commercially, if the licence text and copyright notice travel with the files, which the packs do. The models are unmodified, so no change statement is needed. The packs are separate from the Runtime package.

Bundled in the optional voice-input packs#

Two optional downloads, "MicPack-European" and "MicPack-Asian", hosted on Hugging Face only. Each unpacks into red4ext/plugins/AiNpcCyberpunkMod/app/models/stt/<pack id>/ (beside the Runtime's own Moonshine folder, which it never touches) and carries its model files, a NOTICE.txt with the model's licence and attribution, LICENSE-mod.txt (this project's own licence) and a mic-pack.json marker. The model files are the unmodified files of the sherpa-onnx release archives named below, pinned by size and SHA-256 in packaging/release.config.json (micPacks). The mod uses these models only to turn the player's own speech into text on the player's computer. Neither pack is part of the Runtime package.

• Parakeet (CC-BY-4.0). Attribution is required: NVIDIA Parakeet TDT 0.6B v3, Copyright (c) NVIDIA Corporation, with the licence link and an indication of changes. The changes are the conversion to ONNX and the int8 quantisation done by the k2-fsa sherpa-onnx project (https://github.com/k2-fsa/sherpa-onnx, Apache-2.0); the model is run through sherpa-onnx. NVIDIA does not endorse this mod. The pack's NOTICE.txt carries this text.
• SenseVoice (FunASR Model Open Source License 1.1). The licence lets the model be used, copied, modified and shared; the source and the author must be attributed and the model's name (SenseVoice) kept with it; the model is provided as is, with no liability for Alibaba Group. The licence text is linked, not bundled; the pack's NOTICE.txt names the licence and gives the link. This pack's use in the mod is speech in Mandarin Chinese, Japanese and Korean (the other languages of the model are not used).
• sherpa-onnx itself (Apache-2.0) runs both models; it is in the Runtime package (see the table at the top).

Bundled in the optional PocketTTS model packs#

Five optional downloads, one per language: PocketTTS-Model-100M-French.zip, -Italian, -German, -Spanish and -Portuguese, and six more for Russian, Korean, Japanese, Czech, Turkish and Polish (PocketTTS-Model-100M-Russian.zip and so on, described after the list below), hosted on Hugging Face only (never on Nexus Mods). Each unpacks into red4ext/plugins/AiNpcCyberpunkMod/voices/multilingual-pocket/<lang>/ (a sibling of the voice language pack's multilingual/<lang>/ folder) and carries model.safetensors, tokenizer.json, inference.yaml, a NOTICE.txt with the licence, the attribution, Kyutai's prohibited-use terms and the changes, and a pocket-model.json marker (sizes and SHA-256 of the files). They are built by packaging/build-pocket-models.mjs. The mod uses them only to speak replies in that language, in a character's voice cloned from the game's own dub of that character, with the multilingual pack of the same language. They are not part of the Runtime package.

• CC-BY-4.0. Attribution is required: Pocket TTS by Kyutai (https://kyutai.org/pocket-tts), the licence link, and an indication of changes. Each pack's NOTICE.txt carries all three.
• Changes. model.safetensors and tokenizer.json are byte for byte the upstream files (their SHA-256 values are pinned in the build script and listed in pocket-model.json). inference.yaml is Kyutai's configuration with exactly three values changed, the weight path, the voice-cloning-free weight path (set to null) and the tokenizer path, which point at the local files in the pack instead of hf:// addresses.
• Prohibited use. Kyutai gates the original weights behind an access form that asks for acceptance of its prohibited-use terms, and the model card states them: use "must comply with all applicable laws and regulations and must not result in, involve, or facilitate any illegal, harmful, deceptive, fraudulent, or unauthorized activity", and the prohibited uses "include, without limitation, voice impersonation or cloning without explicit and lawful consent; misinformation, disinformation, or deception (including fake news, fraudulent calls, or presenting generated content as genuine recordings of real people or events); and the generation of unlawful, harmful, libelous, abusive, harassing, discriminatory, hateful, or privacy-invasive content". Kyutai disclaims all liability for non-compliant use. The full wording is in each pack's NOTICE.txt.
• The pack is an optional download from Hugging Face; it does not replace the English Pocket voice model (see "Pocket voice model (Kyutai)" above).

The Russian PocketTTS model pack#

PocketTTS-Model-100M-Russian.zip (folder ru) is a sixth download of the same kind. It replaces the earlier Russian model (xVibePocketTTS by GenVoice, PocketTTS-Model-90M-Russian.zip), whose repository states no licence. It differs from the five above in its source and in three files:

• Attribution (CC-BY-4.0): Pocket TTS by Kyutai (https://kyutai.org/pocket-tts) and the Russian fine-tune Shtorm-PocketTTS-RU by ArtShtorm. The pack's NOTICE.txt names both, links the licences and repeats Kyutai's prohibited-use terms word for word.
• The stress table is an adaptation of CC-BY-SA-4.0 data (the openrussian.org data comes from various sources including Wiktionary and the corrections of its team and users). The word forms were converted into a compact table of plain word to stressed letter, and words with the letter yo, words with one vowel and words with two different stresses in the data were left out. stress-ru.bin is therefore shared under the same licence, CC-BY-SA-4.0, with the source and the build script (packaging/build-russian-stress.py, record in packaging/russian-stress/source.json) in this repository. The licence covers that file and its data, not the mod's code.

The Korean, Japanese, Czech, Turkish and Polish PocketTTS model packs#

Five more downloads of the same kind, PocketTTS-Model-100M-Korean.zip, -Japanese, -Czech, -Turkish and -Polish (folders ko, ja, cs, tr, pl). Each is a community fine-tune of Kyutai's Pocket TTS with six layers, converted to ONNX graphs by scripts/pocket-onnx/export_pocket_onnx.py (the weights are not changed, the tokenizer is converted to tokenizer.json). Each pack's NOTICE.txt names the author, the repository and revision, the licence and Kyutai's prohibited-use terms.

• Attribution (CC-BY-4.0): Pocket TTS by Kyutai (https://kyutai.org/pocket-tts) and each fine-tune by the author named in the row. The Turkish model is MIT: its copyright notice is the author's, named in the pack's NOTICE.txt.
• The Polish model card lists "PWR AZON research recordings" among its training data without stating that data's licence. The weights are published under CC-BY-4.0 and no recording is redistributed. A question to the author about that corpus is open.

Installed separately by players#

Cyber Engine Tweaks, RED4ext, redscript, Codeware and Mod Configuration Menu come from Nexus Mods under their own licences. The Convai service is used under Convai's terms (https://convai.com/tos).