Context Window in AI Chat: Tokens and Limits by App

Quick answer

The context window is the model's working memory: the prompt, your recent chat and the reply must all fit inside it, and text outside it is invisible to the model unless the app puts it back.
It is counted in tokens. By OpenAI's rule of thumb, 100 tokens is about 75 English words.
Published sizes: SpicyChat 4,096 (free) to up to 32,768 tokens; Janitor AI about 9K on the free plan and 5x more on janitor+; CrushOn.AI 8K, 16K or 24K by plan.
Most other apps publish no number. Their memory features (notes, summaries, stored facts) cover what a window cannot hold.

What Is a Context Window?

A context window is the amount of text a language model can consider at one time when producing a reply. Anthropic's documentation describes it as all the text a model can reference when generating a response, "including the response itself", a kind of "working memory" that is separate from the data the model was trained on. Google's documentation compares it to short-term memory.

When a chat grows past the window, the oldest text drops out. That is why a character can forget a name you gave it an hour ago, and why companion apps add separate memory features. Anthropic also warns that a larger window is not automatically better: as the token count grows, accuracy and recall degrade, which it calls "context rot".

Tokens: How Big Is 4K or 24K?

A token is a character, part of a word, a whole word or punctuation. OpenAI says 100 tokens is about 75 words, and Google says a Gemini token is about 4 characters, or 60 to 80 English words per 100 tokens. By that rule of thumb, 4,096 tokens is roughly 3,000 words and 24K roughly 18,000. These are estimates and vary by model and language.

Everything shares the window: character definition, notes, earlier messages and the reply. A long character prompt leaves less room for chat. Do not confuse these tokens with the currency some apps sell: Candy AI and DreamGF use "tokens" as credits, while SpicyChat's docs use the word for text length.

Context Window Sizes by App

App	Context window	Related documented tools
SpicyChat	4,096 free and Get A Taste; 8,192 True Supporter; up to 32,768 I'm All In (model-dependent)	Memory Manager (pin and edit memories); Semantic Memory 2.0 summaries from True Supporter
Janitor AI	About 9K tokens free (the memory article says 8-9K); 5x more on janitor+, no figure published	Chat Memory field you write yourself; the oldest text is forgotten when full
CrushOn.AI	8K Free; 16K Standard, Premium and Luxe; 24K Elite and Imperial	Pinned messages, Auto-Summarization, Save Memories (up to 100 messages)
Character.AI	No figure found	Story Memory, pinned messages, background compression of old context, Memory Usage bar
Kindroid	Not published; higher context on the Ultra and MAX add-ons	Cascaded and retrievable memory, journals
Muah AI	"Membership-based token budget", no figure	Optional Automatic AI Memory Compaction (beta)

Nomi, Replika, Anima, Paradot, Candy AI, DreamGF, EVA AI and GirlfriendGPT publish no context size in the pages we read. Compare memory tools across apps in our ranking of AI companions with memory.

What Happens When the Window Fills Up

Apps handle overflow differently. Janitor AI's help center says the model forgets the oldest text. Character.AI says it compresses old context in the background and shows a Memory Usage bar. SpicyChat's Semantic Memory 2.0 and CrushOn.AI's Auto-Summarization turn older chat into memories. Muah AI warns that "a compact summary is not an archival backup".

If you run a model yourself, you set the window. Ollama's FAQ gives a default of 4,096 tokens, and KoboldCpp says model size and context length decide how much memory you need. See local vs cloud AI companions.

Tip

General advice: keep a character definition short so more of the window holds the conversation. Janitor AI suggests under about 2,500 permanent tokens (best under 1,500); SpicyChat suggests roughly 900 to 1,100. More in memory design for AI characters; other terms are in the glossary.

FAQ

Why does my AI companion forget what I told it?

Usually the detail has left the context window, or the app never stored it as a memory. Pin it, write it into the memory field or add it to the character definition where the app allows.

Does a bigger context window mean better memory?

Not by itself. A larger window holds more recent chat, but Anthropic notes recall degrades as the token count grows, and long-term memory features store facts across sessions.

Is a token the same as a word?

No. A token can be a word, part of a word or punctuation; OpenAI puts 100 tokens at about 75 words.

Sources

Official pages and store listings we read for this page (last check October 20, 2026).