What is a context window?
A context window is the amount of text, measured in tokens, that an AI model can take in at once: the instructions, the conversation so far, any documents, and room for its reply.
Everything a model knows about your request has to fit inside it. What does not fit is, for that reply, not there.
How it works
Text is split into tokens, pieces of a few characters each, and a model's context window is a fixed number of them. Each request sends the relevant history again, so a long conversation, a large file or a big tool result fills the window faster.
When the window is full, something has to give: older messages are dropped or summarized, long results are cut down, or the request is refused.
Examples
- A long chat in which the model seems to forget something said at the start.
- A large PDF that has to be searched in pieces instead of read whole.
- A tool result, such as a long web page, that crowds out the rest of the conversation.
Limitations
- A larger window costs more per request and can be slower.
- Models do not use every part of a long window equally well; details in the middle can be missed.
- A window is not memory: nothing in it carries over to the next conversation by itself.
How Venta uses it
When a conversation grows long, Venta folds the older turns into a short note, so what was said earlier still counts without being sent again. A request still too large for the included model is sent again with long tool results cut down, and Venta says plainly when it is still too large.
For your documents, Venta looks up the passages that match your question and sends those, rather than whole files.
Questions people ask first
- How many tokens is a word?
- In English, a token averages about four characters, so a word is usually one or two tokens. Other languages and code often take more.
- Is a bigger context window always better?
- No. It lets a model read more at once, but costs more and does not guarantee the model uses all of it well. Picking the right passages often works better.
- Does Venta remember things beyond the context window?
- Yes, through memory: what it has stored about you, which you can see, download, and switch off for a project.