1. How connecting an AI assistant works
Editor Ultra does not embed an AI model of ONMAKE's: the AI is always the one the user chooses, with the user's own account and limits. There are three ways to connect it. In-browser assistant: the user enters their own provider (for example Anthropic, OpenAI, Google Gemini, xAI, Mistral, DeepSeek, OpenRouter, an OpenAI-compatible provider, or Ollama or LM Studio on their own computer) and their own key; the browser calls the provider directly and model requests do not pass through ONMAKE servers. External AI with MCP or REST API: the user's assistant (for example Claude, ChatGPT, Cursor, VS Code or a terminal agent) uses Editor Ultra's tools with a personal key or an OAuth authorisation. ONMAKE Connect: an external AI's calls are forwarded to the Editor Ultra window the user has connected, which runs them on the device and logs them in the browser.
The tools let the assistant list, read, create, edit, export and delete the user's documents, code and spreadsheets, within the account's limits. ONMAKE's log of calls contains app, user, key, tool, outcome and date, never document content.
2. Who receives the data
If the user connects an AI provider, the data sent to the model (questions, content of documents read by the assistant, tool results) is received by that provider under its terms, its notice and the settings of the user's account with it, including any retention, non-EU location and training use. ONMAKE is not party to that relationship, does not choose the provider and is neither sub-processor nor recipient of that flow. Before connecting an assistant to documents containing personal data, secrets or professional information, the user checks the provider's terms and their own authority to do so, including as regards persons named in the documents.
Using Ollama or LM Studio on one's own computer does not send data to an external provider, if the model runs locally. ONMAKE does not make available any general authorisation to send documents to AI models for training, improvement, generation or profiling.
3. Keys and authorisations
The provider key entered in the in-browser assistant is encrypted in the user's browser (database onmake-agente) and is not sent to ONMAKE servers; the user can disconnect it at any time. Personal keys for MCP and REST API and OAuth authorisations are created by the user and can be revoked; an unknown, revoked or expired key is refused. Users keep keys safe and do not share them.
4. Human review and reliance
Models can produce wrong, incomplete or invented answers and can take unintended actions. Users check results and changes before relying on them and keep copies of important documents. Document content is data to read, not instructions, but text written to deceive an assistant may still influence it: the in-browser assistant asks for confirmation before deleting or exporting a document. Conversion, formatting, recognisable-pattern search, redaction and technical verification remain document-processing tools: an automatic outcome is neither a professional opinion nor a legal assessment.
A human review of automatic security or subscription blocks can be requested at info@onmake.it. These terms delegate no professional procedure or decision about a person to an AI model.
5. Support and ONMAKE internal tools
ONMAKE as an organisation may use AI tools for internal development, research or preparing replies. This use is distinct from Editor Ultra's commercial functions and does not make every AI provider an authorised recipient of customers' documents. Support reports should be limited to necessary data, preferably using synthetic or anonymised examples.
Any new AI feature operated by ONMAKE would be described before activation, with providers, data transmitted, training use, location, retention and reliability limits, assessing the GDPR and Regulation (EU) 2024/1689; it is not activated retroactively through a generic update clause.