AI transparency

Last updated: 17. Oktober 2026

Osiris is a tool that uses an AI language model. Anyone deploying something like that should know what happens in the process — which model answers, where the input goes and what you must not rely on. This page answers that, regardless of how far the duties of the AI Act reach in any particular case.

1. Who does what here

Osiris is a development environment. It contains no AI model of its own and trains none. The answers come from Claude, a language model by Anthropic PBC, addressed through Anthropic's Claude Code using your own Anthropic access.

The model: Claude by Anthropic PBC. Anthropic's terms of service and privacy notice apply there.
The interface: Colin Pfingsten provides Osiris — the program that calls the model.
The operation: You: Osiris runs on your machine, with your Anthropic access.

This list describes who does what. It is not a legal classification of the roles “provider” and “deployer” within the meaning of the AI Act — that depends on the individual case and remains unsettled for tools that merely operate someone else's model.

2. What data reaches the model

Osiris runs on your machine. Your knowledge graph and its Memory are folders of Markdown files on your disk. Only what a single request needs goes to Anthropic:

your message
the rules and notes Osiris adds from the Ultrabrain
the map of the Memory and the pages the model requests from them
files you explicitly attach

For every project, a context preview shows the rules and notes Osiris hands over, as they go into the CLAUDE.md. Osiris sends your content to us only when you choose to: Social messages, projects you publish in Explore, and feedback. Besides that, our server receives your sign-in, the licence check and, when you open the app, the check for updates and maintenance. More on this under About Osiris.

If you connect ElevenLabs for the voice assistant Jarvis, the text Jarvis speaks goes to ElevenLabs — and, if you choose ElevenLabs for speech recognition in the settings, what you say as well.

3. Labelling generated content

In the chat you are dealing with an AI system; that is evident from the interface at all times. Text, code and files produced this way are machine-generated. If you publish or pass on such content, review it first — and label it where that is called for.

4. Limits — please take them seriously

Language models are convincingly wrong. They produce text that sounds plausible even when it is false. Every output is a proposal, not a result.
The Memory can go stale. What Osiris remembered was true once. The weekly maintenance run looks for contradictions, but it does not find every one.
No legal, tax or medical advice. Osiris is a tool for software development.
You remain responsible. Code that grows out of a proposal is your code. Read it before you ship it.

5. What Osiris does not do

no biometric identification, no emotion recognition, no scoring of people
no automated decisions about human beings
no training of a model on your data
no covert data sharing

6. AI literacy

The AI Act (Regulation (EU) 2024/1689) expects people working with AI systems to understand what those systems do and where their limits lie (Art. 4). Sections 2 to 5 are our contribution to that. Anyone deploying Osiris in a company should pass them on to the people who work with it — and check their own obligations where those apply.

This page informs voluntarily. It replaces no legal review and makes no statement about which duties of the AI Act apply to whom in a given case.

7. Questions

If anything is unclear or you find something missing here, write to support@osirisui.com.