AI Disclosure

Krezl by Treetop Technologies, Inc.

Krezl uses artificial intelligence to help you organize knowledge, structure work, and navigate your information. This document explains how AI is used in the Service, what it can and cannot do, and your responsibilities when using AI-powered features.

1. How AI Is Used in Krezl

1.1 Aria — The AI Assistant

Aria is Krezl's AI-powered assistant. When you interact with Aria, your messages and relevant workspace context are processed by an AI model to generate responses and execute actions within your workspace.

What Aria can do:

• Create, update, and organize entities in your workspace (tasks, projects, features, spaces)
• Search your knowledge graph using semantic and keyword search
• Generate project structures and scaffolds based on your accumulated knowledge
• Retrieve and synthesize information from your workspace
• Set reminders and manage scheduling
• Navigate you to relevant pages in the application
• Conduct web research on your behalf

What Aria does:

• Makes real changes to your workspace (creates tasks, structures projects, etc.)
• Reads your workspace content to provide contextually relevant responses
• Uses your accumulated knowledge (past comments, attachments, decisions) to inform its outputs

1.2 Signal Extraction

When you upload attachments or create comments, the Service may automatically extract structured signals (summaries, key insights, metadata) using AI. These signals are stored within your workspace to improve search and knowledge retrieval.

1.3 Semantic Search

The Service uses AI-generated text embeddings to enable meaning-based search across your workspace. Your content (titles, descriptions, comments) is converted into mathematical representations that allow the system to find related content even when different words are used.

2. AI Models and Providers

3. How Your Data Is Handled

3.1 No Model Training

Your workspace content is not used to train AI models. Content sent to Google Gemini and Vertex AI is processed under Google Cloud's data processing terms, which prohibit the use of customer data for model training.

3.2 Data Minimization

When processing AI requests, we send only the content necessary to generate a useful response — typically your message, the entities you are viewing, and relevant context from your workspace. We do not send your entire workspace to the AI model.

3.3 Data Retention by AI Providers

Under our agreements with Google Cloud, AI providers do not retain your input or output data after processing is complete, except as required for abuse monitoring (retained for up to 30 days and not used for model improvement).

4. Limitations and Risks

4.1 Accuracy

AI-generated content may be:

• Inaccurate — containing factual errors or fabricated information
• Incomplete — missing important considerations or context
• Outdated — based on training data that may not reflect current information
• Inappropriate — not suitable for your specific professional, legal, or regulatory context

4.2 Not Professional Advice

AI-generated content is not professional advice. The Service does not hold any professional license and does not represent itself as qualified to provide legal, financial, medical, engineering, or other professional guidance.

If you use the Service in regulated or professional contexts, you must have AI-generated output reviewed by an appropriately licensed professional before relying on it.

4.3 Probabilistic Output

AI responses are probabilistic. The same question may produce different responses at different times. This is a fundamental characteristic of large language models, not a defect.

5. Your Responsibilities

When using AI-powered features, you agree to:

1. Review before relying. Evaluate AI-generated content for accuracy and appropriateness before acting on it.
2. Maintain professional standards. If you use AI output in professional contexts, ensure it meets the standards of your profession and jurisdiction.
3. Disclose appropriately. Where your professional or ethical obligations require disclosure of AI assistance, make such disclosure.
4. Report issues. If you encounter AI output that is harmful, biased, or significantly inaccurate, report it to support@krezl.com so we can improve the Service.

6. Transparency Commitment

We are committed to being transparent about how AI is used in Krezl. If we make material changes to our AI practices — including changing AI providers, significantly expanding the scope of AI processing, or modifying how your data is handled — we will update this document and notify you.

7. Contact

For questions about AI use in Krezl: ai@krezl.com