AI Transparency

Last updated: April 7, 2026

1. What lynox is

lynox is an AI-powered business assistant that uses large language models (LLMs) developed by Anthropic — specifically the Claude model family — to assist users with business operations. It is developed and operated by Brandfusion Burlet (lynox AI), a Swiss sole proprietorship based in Rapperswil-Jona, Switzerland. lynox is available as a self-hosted open-source application (ELv2 license) and as a managed hosting service.

This document provides transparency about how AI is used within lynox, in the spirit of the EU AI Act (Regulation 2024/1689), specifically the transparency obligations applicable to AI systems interacting with natural persons under Art. 50, as well as the obligations for deployers under Art. 26, and general AI governance best practices.

2. How AI is used

lynox uses AI models for the following purposes:

• Conversational responses — The primary interface is a chat where the AI responds to user queries, drafts content, and provides business advice.
• Memory extraction — The AI identifies and extracts key facts, decisions, and entities from conversations to build a persistent memory layer.
• Knowledge graph construction — Extracted entities (people, companies, projects, topics) and their relationships are organized into a structured knowledge graph.
• Workflow execution — The AI can execute multi-step workflows involving tools (web search, file operations, email, calendar) based on user instructions.
• Content analysis — The AI analyzes documents, emails, and other content provided by the user to extract insights, summarize information, or prepare responses.

3. Human oversight and control

lynox is designed with human oversight as a core principle:

• Permission system — Actions that modify external systems (sending emails, writing files, executing commands) require explicit user approval via the ask_user mechanism. The AI cannot perform write operations without human confirmation.
• Tool pre-approval — Users can configure which tools are available to the AI and set approval requirements per tool category.
• Configurable autonomy — The level of AI autonomy can be adjusted from fully supervised (approval required for every action) to semi-autonomous (pre-approved categories execute automatically).
• Transparent reasoning — Tool calls and their results are visible to the user in the conversation interface.

4. AI limitations

Users should be aware of the following inherent limitations of AI systems:

• Hallucination — AI models may generate information that sounds plausible but is factually incorrect. lynox does not independently verify the factual accuracy of AI-generated content.
• Knowledge cutoff — The underlying AI models have a training data cutoff date and may not be aware of recent events, changes in law, or updated information.
• Confidence calibration — The AI's apparent confidence in a response does not necessarily reflect its actual accuracy. High-confidence responses may still be wrong.
• Context limitations — While lynox augments the AI with memory and knowledge graph context, the model's context window is finite and it may miss relevant information in very long conversations.
• Bias — AI models may reflect biases present in their training data. Users should exercise critical judgment when reviewing AI-generated content.

5. Data and AI processing

How your data interacts with AI models:

• Inference only — Conversations sent to the AI provider (Anthropic or AWS Bedrock) are used for inference (generating responses) only. Neither Anthropic nor AWS use customer data to train or improve their models under their respective API terms of service.
• Local-first storage — All persistent data (conversations, knowledge graph, files, settings) is stored locally on the instance — either on your own server (self-hosted) or on your dedicated managed server.
• Provider coverage — Data sent to the AI provider is covered by the provider's Data Processing Agreement. See our DPA for the full sub-processor list.
• EU processing option — The Managed EU tier routes all AI inference through AWS Bedrock in Frankfurt (eu-central-1), keeping data within the EU.

6. Model information

The same models are available via AWS Bedrock (Managed EU tier) and via direct Anthropic API (Starter tier / self-hosted).

7. No autonomous decision-making

lynox is a tool that assists humans — it does not make autonomous decisions with legal or similarly significant effects. Specifically:

• lynox does not make legally binding decisions on behalf of users.
• lynox does not approve or reject applications, contracts, or financial transactions without explicit human action.
• All AI-generated suggestions require human review and approval before taking effect in external systems.
• lynox operates within a supervised framework where the degree of autonomy is configurable by the user. Even in semi-autonomous mode, tool execution is bounded by safety guards (permission system, cost limits, rate limits) and the user can review all actions taken.

8. No profiling

The lynox knowledge graph captures business entities (contacts, companies, projects) and their relationships to provide context for conversations. This is not personal profiling:

• No scoring, rating, or classification of individuals is performed.
• No automated assessment of personal characteristics, behavior, or creditworthiness.
• No prediction of preferences, interests, reliability, or location of natural persons.
• Knowledge graph data is accessible only to the instance owner and can be reviewed, edited, or deleted at any time.

9. User rights regarding AI-generated content

Users have full control over all AI-generated content within their lynox instance:

• Delete conversations — Any conversation and its associated AI-generated content can be permanently deleted.
• Disable memory extraction — Memory extraction can be turned off, preventing the AI from persisting information between conversations.
• Review knowledge graph — All entities and relationships in the knowledge graph can be viewed, edited, or removed.
• Export data — All data can be exported in standard formats.
• Full deletion — Upon account termination (managed hosting), all data is permanently deleted as specified in our DPA.

10. AI Act classification

Based on our assessment, lynox falls under the category of general-purpose AI systems used for business productivity and is not classified as a high-risk AI system under Annex III of the EU AI Act. lynox does not operate in any of the high-risk domains enumerated in the regulation (critical infrastructure, education, employment, law enforcement, etc.).

This transparency document is provided voluntarily as a matter of responsible AI governance and to meet the transparency obligations under Art. 50 (interaction with natural persons) and Art. 26 (obligations for deployers) of the AI Act.

11. Contact

For questions about AI use in lynox, data protection, or this transparency document:
[email protected]