AI Transparency

TokTalk uses artificial intelligence to convert your speech into polished text. In the spirit of the EU AI Act and our commitment to transparency, this page explains exactly which AI systems we use, what they do, and—critically—what they do not do.

1. AI Systems We Use

TokTalk employs two AI models, both developed by Mistral AI, a European AI company headquartered in Paris, France:

• Speech-to-Text (Voxtral Mini): Converts your spoken audio into raw text. This model listens to your voice and produces a verbatim transcription of what you said.
• Text Enhancement (Mistral Small): Improves the raw transcription by correcting grammar, removing filler words (like "um" and "eh"), and polishing sentence structure—according to your chosen enhancement level and writing style.

2. What AI Does in TokTalk

• Converts spoken language into written text.
• Removes verbal tics, stuttering, and filler words.
• Corrects grammar, punctuation, and capitalization.
• Restructures sentences for clarity (at higher enhancement levels).
• Respects your custom dictionary terms and text snippets.

3. What AI Does NOT Do

We believe in clear boundaries for AI. TokTalk's AI explicitly does not:

• Make decisions on your behalf. The AI only formats text—it never takes autonomous actions, sends messages, or makes choices for you.
• Add information. The AI never invents facts, names, numbers, or URLs that you did not speak.
• Answer questions. If you dictate a question, it remains a question in the output. The AI does not attempt to answer it.
• Profile or score users. We do not use AI to analyze user behavior, create profiles, or assign any kind of risk or credibility scores.
• Train on your data. Your audio, transcriptions, and personal data are never used to train or improve AI models—neither ours nor our providers'.

4. Risk Classification

Under the EU AI Act, AI systems are classified by risk level. TokTalk falls under the "limited risk" category because:

• It is a productivity tool, not a system that makes consequential decisions about people.
• It does not operate in any high-risk domain (healthcare, law enforcement, education, employment, etc.).
• It does not use biometric identification or emotion recognition.
• Users always retain full control over the AI output and can edit or discard it before use.

As a limited-risk AI system, our primary obligation is transparency—which this page fulfils.

5. Human Oversight

TokTalk is designed as a human-in-the-loop system. The AI generates a text suggestion, but you always see the output before it is used anywhere. You can edit, accept, or discard the transcription entirely. The AI never acts autonomously—it is a writing assistant, not an agent.

6. Data Protection

Audio recordings are processed in transient memory and never stored on disk. Resulting transcriptions are stored in your account for history purposes and can be deleted at any time. Neither audio nor text is used for AI model training. For full details, see our Privacy Policy and GDPR Compliance page.

7. Contact

If you have questions about our use of AI, please contact us at:

Email: info@toktalk.co