You do not need any of this to start. Skim it once, then come back when a word has appeared in three conversations and you have decided it is not going away.
Four terms do most of the work in everyday conversation here, and they are worth knowing early: token, context window, hallucination, and harness. If you learn only four words from this page, learn those.
- Agent
- A model operating through a harness toward a goal, often using tools and taking multiple steps.
- AI model
- A trained computational system that turns inputs into outputs such as text, images, classifications, or predictions.
- API
- A defined way for software systems to exchange requests and data.
- Branch
- A separate line of development that lets you experiment without immediately changing the main version.
- Commit
- A saved snapshot of a set of changes in version control.
- Context window
- The amount of information a model can consider in one interaction or working context.
- Data sovereignty
- The ability to control where data is stored, how it is governed, and which systems it depends on.
- Deployment
- Moving a tested change or application into the environment where it will run.
- Development environment
- The place where software is built, tested, and changed. Safe to break.
- Git / version control
- A system for recording changes so you can compare, collaborate, and return to earlier versions.
- Hallucination
- A model output that sounds plausible but is unsupported, incorrect, or invented.
- Harness
- Software that surrounds a model with instructions, files, memory, tools, permissions, and an interface.
- Inference
- Using a trained model to generate an answer or prediction.
- LLM
- Large language model: a model trained to work with language and related structured information.
- Local model
- A model running on your own computer or infrastructure rather than only through a remote provider.
- Model provider
- A company or service that hosts models and provides access to them.
- Multimodal
- Able to work with more than one type of input or output, such as text, images, audio, or video.
- Open source
- Software whose source code is available under a licence that permits inspection, use, and modification.
- Open weights
- A model whose trained parameter files are available. This does not automatically mean every part is open source.
- Production environment
- The live system used by real users or operations. Changes here should be deliberate and recoverable.
- Prompt
- The instructions and context you give a model.
- RAG
- Retrieval-augmented generation: retrieving relevant information first, then giving it to a model as context. The usual way to make an assistant answer from your own documents.
- Repository / repo
- A project folder whose files and revision history are managed together.
- Terminal / CLI
- A text-based interface for giving commands directly to a computer or program.
- Token
- A unit a model uses to process input and output. It may be part of a word, a whole word, or punctuation. See how the pieces fit together for a worked example.
- Tool calling
- A model requesting that software perform an external action, such as a search, code execution, or an API call.
Reference points for two of the above, if you want the primary source: Pro Git on version control, and OpenAI on tokens.
A word that is missing and keeps coming up in the Discord is a reasonable thing to post there. A glossary that grows from actual confusion is more useful than one that starts complete.