Context window
The context window is the maximum amount of text, counted in tokens, that a language model can consider at one time. It holds your instructions, the conversation so far, any documents you paste and the reply being written.
Last updated November 8, 2026
Readable on your devices only.
Your conversation is readable on your device, and on your other devices if you sync them. Nowhere else.
Unreadable to us.
Your message passes through without being written down; your chats are only ever stored encrypted.
The model doesn’t know you.
It gets the question, never your email, your account or your IP address.
Nothing is left along the way.
The reply comes back and is stored here, encrypted; even when you sync, your key never reaches us in the clear.
How it works
A model has no memory between requests. Each time you send a message, the app sends the whole conversation again, and the model reads it from the start. Everything that fits in the window is visible to the model; anything outside it simply does not exist for that answer. The window is shared between what goes in and what comes out, so a long prompt leaves less room for the reply, and many models also set a separate ceiling for the length of one reply. Image attachments, tool definitions and hidden instructions occupy space as well. Think of a desk: a small one forces you to file papers away, a big one holds the entire case.
A bigger window lets you work with longer material: a contract, several chapters of a manuscript, a meeting transcript or a large part of a code base, without cutting it into pieces: a forty-page contract, a quarterly earnings call, a thesis draft, an exported spreadsheet or a whole repository. It has costs, though. Reading more tokens takes more time and, with paid models, more money. Models can also handle very long inputs unevenly, and details buried in the middle of a huge text are sometimes missed, so a large window is not a guarantee that every line is used well.
When a conversation outgrows the window, the app has to choose. It can drop or shorten the oldest messages, replace them with a summary, or ask you to start again. Some apps keep the opening instructions and the latest turns and condense whatever lies between. Another approach is retrieval: instead of pasting a whole library, you search it and give the model only the passages that bear on the question. Architectural tricks such as sliding-window attention, rotary position embeddings and key-value caching let engineers stretch windows to hundreds of thousands of positions, although memory consumption and latency still climb as the prompt lengthens.
In NeverNo
Each model on NeverNo has its own window, shown on its page. Most of the catalog offers 128,000 or 200,000 tokens, Claude Haiku 5.5 has 100,000, and Kimi K3, the private reasoning model from Moonshot, has the largest at 256,000 tokens. That is enough to put a long report or a stack of documents into one conversation and ask questions about the whole of it.
The page about the ChatGPT alternative describes the same idea for people moving from another assistant: bring a long document, keep the conversation, and choose the model whose window fits the job. For collections that are too big for any window, the Vault keeps your documents encrypted in your browser and sends only the passages a question needs.
Worth remembering
The window is counted in tokens and covers instructions, history, documents and the reply together.
A model sees nothing beyond its window, so older content must be shortened, summarized or retrieved.
A larger window is useful for long material but is slower and can still overlook details in the middle.
The size differs by model: on NeverNo it ranges from 100,000 to 256,000 tokens.
Output counts too: a prompt that nearly fills the window leaves almost no space for the answer.
Frequently asked questions
Is the context window the same as the model’s memory?
Not quite. It is the working space for one request. The model does not keep anything afterward; the app resends the conversation each time. Long-term memory features are separate and only add short notes into that same window.
Does a model use the whole window well?
Not always. Very long inputs can dilute attention, and a fact placed in the middle may be overlooked. Putting the key question and the essential passages clearly, and asking for quotations, helps. Needle-in-a-haystack benchmarks probe exactly this weakness.
What should I do with a document larger than the window?
Split it, summarize part by part, or use retrieval: index the document, search for the passages related to your question and give the model only those. The retrieval-augmented generation entry explains the method. Chunking by chapter, section headings or timestamps keeps each piece coherent when you must split material.
Ask your first question
No account and no card. Write in any of the 10 languages NeverNo speaks.
Open the chat