Date: 2026-05-26
Version: 1.2
Provider: KIMIKON e.U., Amalienstraße 29, 1130 Vienna, Austria
Contact: support@kimikon.cc
1.1. KIMIKON grants you a non-exclusive, non-transferable right to use KimiTalk on the Macs covered by your license tier.
1.2. KimiTalk is available as a free "Basic" tier and a paid "Pro" tier (subscription or lifetime). Tier scope is described in-app and at https://kimikon.cc/kimitalk/.
1.3. KimiTalk is distributed via a signed, notarized DMG and Sparkle auto-updates.
2.1. A purchased license binds to a specific device via a license key.
2.2. A license may be deactivated and reactivated on another device, subject to tier limits.
2.3. KIMIKON may revoke a license obtained or used fraudulently.
KimiTalk converts speech to text and may optionally rewrite or enhance text via local AI language models. The AI models run on-device on your Mac. Dictation content, local text enhancement, and local AI Chat inference are not transmitted to KIMIKON, Apple, Google, Alibaba, Meta, or other model providers. However, if you enable or trigger optional AI Chat web research, search queries and selected webpage fetches are transmitted to external search or website providers; see Data Protection (§ 6 of this EULA and the Privacy Policy).
models.kimikon.cc; KimiTalk does not mirror public argmaxinc/whisperkit-coreml artefacts for the product path.AI-generated content may be inaccurate, incomplete, or inappropriate. You are solely responsible for reviewing AI output before publication, transmission, or use in official documents. Neither KIMIKON nor Apple, Google, Alibaba, Meta or OpenAI warrant correctness, completeness, lawfulness or fitness for any particular purpose.
KimiTalk is not medical, legal, financial or governmental professional advice and is not intended for high-risk decisions. You undertake not to use KimiTalk for prohibited, unlawful or high-risk purposes, including regulated healthcare, legal or financial services, sensitive biometric inference, social scoring, or law-enforcement purposes.
All transcripts and AI outputs created with KimiTalk belong to you. Neither KIMIKON nor any underlying model provider claims ownership of your outputs.
4.1. You undertake to use KimiTalk only in compliance with applicable law. In particular you are responsible for:
4.2. KimiTalk is not approved as a sole documentation tool for high-risk domains (healthcare, legal, financial) and may only be used as an aid alongside human review.
KimiTalk integrates open-source software. See Settings -> "Licenses & Model Notices" and the NOTICE.md file in the app bundle / https://kimikon.cc/kimitalk/notice/. The respective licenses (MIT, Apache 2.0, BSD-2-Clause, Zlib, etc.) take precedence to the extent they contain more specific terms.
6.1. Dictation content, voice embeddings, local text enhancement, and local AI Chat inference are generally processed locally on your Mac.
6.2. KimiTalk downloads release model files and model-support files through KIMIKON's model mirror models.kimikon.cc (Cloudflare R2).
6.3. KimiTalk checks for updates via Sparkle; app version and technical headers are transmitted to updates.kimikon.cc.
6.4. If you enable or trigger optional AI Chat web research, KimiTalk may locally create or rewrite a search query from your message and, where applicable, the chat context, transmit that query to DuckDuckGo HTML Search, and fetch selected result webpages. These queries and fetches leave your device and may contain personal, confidential, or professional-secrecy-relevant content if you enter such content into the chat. Do not use web research for content that should not be transmitted to external search or website providers. To be confirmed by counsel.
6.5. KIMIKON does not collect dictation content, voice embeddings, or personal text input for its own server-side processing. For optional AI Chat web research, however, the external recipients described in § 6.4 and in the Privacy Policy apply.
6.6. Further details: https://kimikon.cc/kimitalk/privacy/en.
7.1. KimiTalk is provided "AS IS" and "AS AVAILABLE".
7.2. KIMIKON has unlimited liability under Austrian law for injuries to life, body, or health, for intent and gross negligence, and under the Product Liability Act.
7.3. In cases of slight negligence, KIMIKON's liability towards consumers is excluded or limited only to the extent such exclusion or limitation can be validly agreed under applicable Austrian consumer law, in particular § 879 ABGB and § 6 KSchG. To the extent a limitation of liability is permissible, liability is limited to the typically foreseeable damage.
7.4. Other liability is excluded to the extent permitted by law. In particular, KIMIKON is not liable for indirect damages, lost profits, data loss, or damages arising from erroneous AI output.
7.5. Mandatory consumer protection rights under Austrian KSchG and EU law remain unaffected.
EU consumers have a statutory right of withdrawal of 14 days. For digital content the right expires once you expressly consent to performance before the withdrawal period ends and acknowledge loss of the right (Art. 16(m) Consumer Rights Directive). KIMIKON will provide you, after contract conclusion, with confirmation of the contract and of your consent/acknowledgement on a durable medium (e.g. by email). Withdrawal: support@kimikon.cc.
9.1. Pro subscriptions auto-renew unless cancelled before period end.
9.2. Pro Lifetime licenses are one-time purchases for a specific major version per tier description.
9.3. KIMIKON may terminate for material breach (in particular §§ 3.3, 4, 11-20). In case of justified termination due to a culpable material breach, no refund is owed to the extent legally permissible and to the extent the service has already been provided. Mandatory consumer rights, statutory refund claims, and a proportional assessment for service periods not yet provided remain unaffected.
10.1. Governing law: Austrian law, excluding conflict-of-laws and CISG. Consumer protection at habitual residence remains unaffected.
10.2. Venue: for entrepreneurs, Vienna. For consumers, statutory venue.
10.3. Severability applies.
10.4. KIMIKON may amend this agreement with 30 days' notice. Subscribers may extraordinarily terminate.
11.1. KimiTalk provides Text-to-Speech ("TTS") with the following engines:
AVSpeechSynthesizer.11.2. Release open-source and community models (b)-(e) are prepared through KIMIKON's model mirror models.kimikon.cc; model weights and helper files are locally verified against signed manifests and SHA-256 checksums. A model-dependent feature is enabled in the product build only when the required KIMIKON mirror is live, hash-verified, and wired into the product.
12.1. Apple System Voices (Standard, Enhanced, Premium, Personal) are governed by the macOS Software License Agreement (Apple Inc.), which restricts use to personal, non-commercial use only. Apple Inc. expressly prohibits:
"No other creation or use of the System Voices, Live Captions or Personal Voice is permitted by this License, including but not limited to the use, reproduction, display, performance, recording, publishing or redistribution of any of the System Voices, Live Captions or Personal Voice in a profit, non-profit, public sharing or commercial context." (macOS Software License Agreement, Apple Inc., 2026)
12.2. KimiTalk does not provide Apple System Voices as its own voice model and does not market them as a Pro voice feature. Apple Speech is an integration with the macOS system functionality available on your Mac. KIMIKON e.U. does not grant you any rights in Apple System Voices.
12.3. KimiTalk does not create exportable WAV files using Apple System Voices. WAV export and batch export are provided through local add-on models where available in your tier and installed on your Mac.
12.4. Personal Voices (self-trained voices on macOS 14+) are filtered out of the Voice Picker. If they nonetheless appear, the obligations of §§ 12.1-12.3 apply in a stricter form (Apple permits "you using your own" only).
13.1. Models (b)-(d) are under Apache License 2.0. Full license texts in LocalSpeechHelper.app/Contents/Resources/NOTICE.txt and at https://kimikon.cc/kimitalk/speech-helper/NOTICE.txt.
13.2. KIMIKON e.U. is not the provider of these models and disclaims warranty for content quality or availability.
13.3. Models run on-device. No transmission of inputs/outputs to Alibaba Cloud, hexgrad, or third parties.
Notice: Per Boson Higgs Audio 2 Community License §§ 1.b.i and 1.b.iii the following attribution must remain verbatim English. A German rendering would be informative only.
14.1. The OmniVoice engine may include the Higgs Audio v2 Tokenizer by Boson AI USA, Inc. (bosonai/higgs-audio-v2-tokenizer) as a runtime component. The tokenizer is licensed under the Boson Higgs Audio 2 Community License Agreement. That license incorporates the Meta Llama 3 Community License and describes the "Higgs Materials" as derived from Meta Llama 3; the resulting Boson/Meta obligations are passed through in this clause.
14.2. Verbatim attribution (mandatory, Boson License § 1.b.i):
Built with Higgs Materials licensed from Boson AI USA, Inc., Copyright Boson AI USA, Inc., All Rights Reserved and Meta Llama 3 licensed under the Meta Llama 3 Community License, Copyright Meta Platforms, Inc., All Right Reserved". based on Meta Llama 3”
14.3. Verbatim notice (mandatory, Boson License § 1.b.iii):
“Meta Llama 3 is licensed under the Meta Llama 3 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved.”
“Boson Higgs Audio 2 is licensed under the Boson Community License, Copyright © Boson AI USA, Inc. All Rights Reserved.”
14.4. Meta Llama 3 Acceptable Use Policy — pass-through (Boson License § 1.b.iv). By activating OmniVoice you undertake to comply with the Meta Llama 3 Acceptable Use Policy as currently available at https://www.llama.com/llama3/use-policy. Prohibited in particular:
14.5. No model improvement (Boson License § 1.b.v): You shall not use the Higgs Materials or any output thereof to improve any other large language model (except Boson Higgs Audio 2 or its derivatives).
14.6. Annual active users cap (Boson License § 2): The Boson Community License permits exercise of the license rights only while the products or services made available by KIMIKON e.U. and its affiliates did not exceed 100,000 annual active users in the preceding calendar year. If that aggregate population is exceeded, KIMIKON is not authorized under Boson License § 2 to exercise rights under the Boson license unless or until Boson AI expressly grants an expanded license. KIMIKON will therefore request an expanded license with a sufficient safety margin before the threshold is reached and will pause or disable OmniVoice in whole or in part before exceeding the threshold if no expanded license has been granted by then. If such disablement materially affects a paid Pro feature during an ongoing paid period, KIMIKON will provide, as required by mandatory consumer law, an appropriate substitute function, a special termination right, or a pro-rata refund for the period during which the feature can no longer be used.
14.7. Update heartbeat / active-install estimate: KIMIKON uses anonymous Sparkle update-check frequency (unique installation identifier, 30-day rolling window) as an early-warning signal for active KimiTalk installations and for internal annual-active-user risk assessment. For Boson License § 2, KIMIKON evaluates the aggregate population of the products or services made available by KIMIKON e.U. and its affiliates. As long as KimiTalk is KIMIKON e.U.'s only publicly provided product/service and no affiliates are added, the KimiTalk signal equals that aggregate population; otherwise additional active users will be included in the internal compliance assessment, or OmniVoice will remain disabled until a reliable aggregate assessment or expanded license exists. No OmniVoice-specific telemetry, advertising tracking, or content analytics are collected on user devices. Aggregated values are checked internally at least monthly against warning, safety-margin and license thresholds.
14.8. OmniVoice code/model: Apache 2.0, Copyright (c) 2026 Xiaomi Corp. The OmniVoice config uses a Qwen3ForCausalLM architecture; the model card names Qwen/Qwen3-0.6B as its basis. Qwen3-0.6B is Apache 2.0, Copyright Alibaba Cloud Qwen Team. The license notices are included in NOTICE.txt.
14.9. Jurisdiction note: The Boson Higgs Audio 2 Community License specifies California law, exclusive venue in the Northern District of California (Santa Clara County). This clause governs the KIMIKON-Boson relationship, not the KIMIKON-user relationship; § 10 of this EULA (Austrian law) applies between you and KIMIKON.
15.1 KimiTalk enables creation of a synthetic voice profile from your reference audio (3-15 seconds).
15.2 You represent and warrant that the reference audio is either:
15.3 You shall indemnify and hold harmless KIMIKON e.U. against any third-party claims arising from breach of § 15.2, including:
15.4 When publishing AI-generated audio recognisable as the voice of an existing person, you must label it as AI-generated per EU AI Act Art. 50(4). KimiTalk provides non-disableable technical marking: filename suffix -ai.wav, BWF-bext metadata in the WAV header, and mandatory AudioSeal watermarking plus C2PA manifests for helper-backed AI audio. If the mandatory provenance layers cannot be created, AI-audio WAV export fails.
15.5 The following are prohibited in particular:
15.6 In case of breach KIMIKON may terminate the voice-cloning license and disable the affected profile.
16.1 Speaker embeddings are considered biometric data under GDPR Art. 4(14) and subject to Art. 9(1).
16.2 Processing occurs exclusively on-device. KIMIKON receives neither reference audio nor embedding nor generated outputs.
16.3 Legal basis is explicit consent under Art. 9(2)(a) GDPR, given in the in-app consent flow before first voice profile creation (biometric processing, AI marking, anti-misuse, and self-attestation of the voice source).
16.4 You may delete your voice profile any time via Settings. Deletion irrevocably removes embedding and reference audio.
17.1 All audio content produced by KimiTalk is AI-generated. It is not spoken by real persons even if it resembles a reference speaker.
17.2 AI-generated speech may contain unexpected pronunciation, prosody or acoustic artifacts. Review audio before any distribution.
17.3 For safety-critical, legally binding or medical use you assume sole responsibility; KIMIKON is liable only for proven gross negligence.
18.1 KimiTalk is a Limited-Risk AI system under EU Regulation 2024/1689 (transparency obligations, Art. 50).
18.2 KIMIKON e.U. implements provider obligations under Art. 50(2) through:
-ai.wav as human-readable marking;18.3 You as "Deployer" (Art. 50(4)) are obliged, when distributing audio recognisable as the voice of an existing person, to clearly disclose the AI generation at the latest at the time of first interaction or exposure.
The Speech Helper DMG bundles libsndfile under LGPL-2.1+. The source code of the bundled version is available on written request to legal@kimikon.cc for a fee not exceeding the cost of physical distribution.
20.1 Mandatory consumer rights under Austrian KSchG/ABGB remain unaffected.
20.2 KIMIKON e.U. excludes liability for damages arising from TTS / voice cloning to the extent caused by:
20.3 The liability rules in § 7 also apply to TTS / voice-cloning features. For consumers, liability for slight negligence is excluded or limited only to the extent permitted under applicable Austrian consumer law, in particular § 879 ABGB and § 6 KSchG; mandatory statutory liability remains unaffected.
20.4 Intent, gross negligence, and mandatory statutory liability remain unaffected.
Contact: KIMIKON e.U., Amalienstraße 29, 1130 Vienna, Austria, support@kimikon.cc, https://kimikon.cc/kimitalk/
Appendix A: Open-source notices in NOTICE.md / https://kimikon.cc/kimitalk/notice/.