Glossary

The vocabulary of sovereign document AI.

Short, plain-language definitions: each term in one place, cross-linked to the wiki.

• Agentic retrieval

  An agent that searches files on demand instead of one-shot vector lookup.

• BYOK (Bring Your Own Key)

  Use your own model API key: data and billing stay on your account.

• CLOUD Act

  A US law that reaches data held by US-owned providers, wherever it sits.

• Context window

  How much text a model can consider at once, measured in tokens.

• Data residency

  Where your data physically sits: the weakest of the three guarantees.

• Data sovereignty

  Which laws govern your data, not just where it sits.

• Deterministic AI

  Letting code compute exact facts so the model only reports them.

• Digital sovereignty

  Control over your whole stack: software, operations, and supply chain.

• Embedding

  A numeric vector representing a chunk of text, the basis of vector search.

• EU AI Act

  The EU's phased AI regulation: self-hosting makes you a deployer, not exempt.

• Frontmatter

  A small YAML metadata block at the top of a markdown file.

• Inference vs. training

  Training builds a model once; inference is running it to get answers.

• MCP (Model Context Protocol)

  A standard interface for agents to talk to tools and data sources.

• OKFZ

  pdf2okf's portable, shareable knowledge bundle.

• On-device / local AI

  Running inference on your own machine: data never leaves.

• Open Knowledge Format (OKF)

  Google's vendor-neutral standard: AI knowledge as markdown + YAML frontmatter.

• Open weights

  Downloadable model weights you can run yourself, not always open source.

• Quantization

  Shrinking a model by storing its weights at lower precision.

• RAG (Retrieval-Augmented Generation)

  Feeding a model retrieved document passages to ground its answers.

• Vector database

  An index of embeddings used by classic RAG, which pdf2okf doesn't need.