Tokens: the pieces a model processes
A token is a piece of encoded input or output. In text, it might be a word, part of a word, punctuation, or whitespace. Different models use different tokenizers, so there is no universal word-to-token conversion. A longer prompt or answer usually uses more tokens.
The context window is the working capacity for information a model can consider in a request. Exactly what counts against it depends on the system. A huge context window does not guarantee that a model will use every detail accurately.
Handles: names or references, depending on context
A handle is not a standard ingredient of every AI system. On a social platform it can be an account name. In software it can be a reference used to access an object, file, or session. In a model interface, people may informally mean a model identifier. Ask where the term appears.
Do not confuse a handle with a harness. If someone is describing the software that lets an AI use tools and keep working, they probably mean harness. A model ID such as the identifier shown in a local app is a name that selects a model; it is not the whole agent.
Harnesses: the surrounding machinery
A harness connects a model to a task. It can supply instructions, tools, a record of progress, and rules for when to stop or ask. Think of a workbench: the model is one component; the workbench determines what tools are within reach and how their results come back.
A practical reading habit
When a headline uses an unfamiliar word, replace it with the specific behavior. Does “memory” mean saved preferences or the current chat? Does “agent” mean it can send an email or only draft one? Specifics make feature comparisons easier and reduce expensive misunderstandings.