Your AI SaaS business, ready by this week!
Try demoYour AI SaaS business, ready by this week!
Try demoThe context window is the maximum number of tokens a model can take into account at once, covering the instructions, the conversation, any attached documents and its own answer.
Everything the model needs for one answer has to fit in the context window: the system prompt, earlier messages, files, search results and the reply itself. Anything that does not fit is cut or never seen.
Context windows have grown from a few thousand tokens to hundreds of thousands and, for some models, over a million. A bigger window lets a model read whole reports or codebases in one go.
A bigger window is not free. Every token in it is billed on every request, and quality can drop when the relevant part is buried in a very long input. RAG and summarising old messages keep the context small and focused.
Aikeedo sends the conversation, memory and attached files within each model’s limits, and uses retrieval for knowledge bases so only the relevant passages take up context.
Discover more terms connected to this one
Tokens are the small pieces of text an AI model reads and writes, roughly four characters or three quarters of an English word each, and the unit most AI providers bill by.
An LLM (large language model) is an AI model trained on very large amounts of text to predict the next token, which lets it write, summarise, translate, reason and hold a conversation.
RAG (retrieval-augmented generation) is a technique where relevant passages are first retrieved from a knowledge base and then given to an AI model with the question, so the answer is grounded in that content.