Chat & AI
When a chat gets too long for its model
What Smart Context, the context limit card and the Switch Model button do, and when to start fresh.
Step 1
Every model has a limit
A model can only read so much at once: the conversation, the files you attached and its own reply. Long chats with big documents get there eventually. Each model's window is shown in the model picker, and Secretary warns you before it becomes a problem.
Anthropic
OpenAI
GoogleStep 2
Smart Context keeps you going
When the conversation no longer fits in full, a violet Smart Context chip appears under the composer. From then on each message is sent with the parts of the history that matter to it, instead of all of it, so you can keep using any model. Hover the chip for the explanation.
Press Enter to send, Shift+Enter for new line
Smart ContextStep 3
The context limit card
As you approach the limit, a card in the conversation says how much of the window is used: amber while approaching, red once it is almost or fully reached. If your attached files alone are too large for the model, it says that instead.
GPT-5.4 Mini is approaching its context limit
This conversation has used 82% of GPT-5.4 Mini's context window. You can continue, switch models for more headroom, or start a fresh chat.
82% of context used
Step 4
Switch Model
Switch Model on the card lists models with more headroom and greys out the ones that would not help. In the model picker, a model that is too small for this chat is greyed out with Context too large. Switching only affects your next message.
GPT-5.4 Mini is almost at its context limit
This conversation has used 96% of GPT-5.4 Mini's context window. You can continue, switch models for more headroom, or start a fresh chat.
96% of context used
Step 5
Or start fresh
Start New Chat on the card opens a new chat with a short summary of this one carried over, so you keep the gist without the weight. In a hub, Start New Hub is offered too.
GPT-5.4 Mini reached its context limit
This conversation is too long for this model to respond. Switch to a model with a larger context window to continue, or start a fresh chat.
Step 6
If a reply fails
A failed reply shows a window naming the cause, such as Conversation Too Long, Rate Limit Reached or AI Service Unavailable. Most offer Switch Model: pick another model and the message is sent again for you. Out of credits points you to Usage or Billing instead.
Conversation Too Long
GPT-5.4 Mini has reached its context limit
This conversation is too long for GPT-5.4 Mini to process. Switch to a model with a larger context window (like Claude) or start a fresh chat.
Try it in Secretary, or create an account to get started.
Related guides
Starting a new chat
The start screen, starter prompts, temporary chats, and what New chat does once you are already in a conversation.
Attaching files to a chat message
Add a file from your computer so the AI can read it: press the paperclip, drag it in, or paste it.
Scoping a chat to specific documents or folders
Point the AI at exactly the files in your library it should answer from, and choose how deeply it reads them.