Glossary
- Agent
- A model that doesn’t just answer but carries out steps itself: creating files, sorting mail, operating programs.
- Approval
- A person’s explicit OK for one specific step by the AI. A “yes” to several questions at once is not an approval but a guess.
- Checksum
- A string of characters calculated from a file’s content. If two checksums match, the files are identical – byte for byte.
- Connector
- A ready-made integration you activate in the AI app with a few clicks so the AI can access a program or service.
- Cross-check
- A second model checks the first model’s result – without its conversation history, against a fixed list. It doesn’t know the first model’s intentions and therefore finds what the first one overlooks.
- Data processing agreement (DPA)
- The contract under Art. 28 GDPR that you need when a service provider processes personal data on your behalf.
- Local model
- A model that runs on your own device, for example with Ollama or GPT4All. Nothing leaves the house, but it is considerably weaker than the big services.
- Markdown
- A simple text format that uses a few characters for headings and lists. Every program can read it; you aren’t tied to any vendor.
- MCP
- Model Context Protocol: an open standard for connecting a model to mail, calendar, files or other programs – like a standard for power outlets.
- Model
- The program that understands and generates text – such as Claude, GPT or Gemini. Often simply called “AI” or “language model.”
- Net / gross
- Net is the price without VAT, gross includes it. Providers show sometimes one, sometimes the other.
- Prompt
- The instruction you give a model. A good prompt says what the model should do and what it should leave alone.
- Scheduled task
- An AI run that starts at a fixed time without anyone doing anything, for example every morning.
- Team vault
- A shared knowledge store into which team members deliberately place part of their own vault. Your own vault remains the authoritative source.
- Thread
- A single conversation with the AI, with its own history.
- Training
- How a model learns from large amounts of text. If your chats are used for training, their content can help shape future models.
- Vault (knowledge store)
- A folder of plain text files on your own computer that the AI reads and writes. Its memory beyond the single conversation.
- Working memory (context)
- What a model has in front of it at once during a conversation. When it fills up, the system summarizes older material – and instructions can get lost in the process.