9.1 What is a context window and why is it important?
It is the maximum amount of text (in tokens ≈ 0.75 words) that an AI model can process simultaneously. In case of saturation, the model "forgets" the initial instructions and may hallucinate. To remedy this: fragment complex tasks into sub-steps, avoid sending long useless histories, and store knowledge in a database rather than in context.
9.2 What is an AI hallucination and how do I avoid it?
A hallucination occurs when the AI invents information due to insufficient context or context window saturation. Solutions: use low temperatures, rigorously structure your data, fragment complex requests, and validate outputs before using them.
9.3 What is Few-Shot Prompting and what is it for?
Few-Shot Prompting consists of giving the AI 2 or 3 examples of your writing style or way of working so that it adopts your stylistic patterns. Result: less generic content and closer to your voice.
9.4 What is the difference between using AI via a chat interface and via API?
The Chat interface (e.g., ChatGPT) is a consumer product with native conversation history management, but data transits through the providers' servers. The API gives access to the raw power of the model, bills per request, and gives the creator (via Celestory) full control over the history and system prompt — ideal for custom business applications.
9.5 Does artificial intelligence store the confidential data of my clients/learners?
This depends on the AI engine (LLM) invoked. Some software has the right to study your content to optimize their models. However, via API requests (especially with European companies like Mistral AI and their Zero Data Retention standard), the machine simply calculates your response and then deletes all memory — categorically preventing any data leak. This is ideal for legal, medical, or corporate training environments.
9.6 Who owns the works generated by artificial intelligence in my modules?
A visual, audio or element generated 100% by AI does not possess exclusive authorship (an AI has no legal personality). However, the logical structure, the educational methodology, and the line of your exclusive prompts created in your Graph or Master Prompt constitute your exclusive Intellectual Property.
9.7 How do I call an external API that has no dedicated integration (not Airtable, Baserow, or Zapier)?
Use the HTTP Request block, which lets you call any URL directly with the GET, POST, PUT, DELETE or PATCH methods, configuring the headers and request body yourself. It is the most flexible block for connecting to a third-party service without a preconfigured integration.
9.8 What is the MCP protocol and what is it used for in Celestory?
MCP (Model Context Protocol) is a standard that lets an AI model call external tools itself during a conversation (querying a database, triggering an action) instead of just generating text. In Celestory, the MCP block dynamically exposes the tools of an MCP server configured on your project, both in the graph and in the AI Generation tab's chat — the model decides on its own when to use them.