The short answer
GDSense keeps extended chats focused by summarizing older messages once a conversation nears the model’s context window, while your most recent exchanges stay word for word. The summary is saved with the conversation and extended as the chat grows, so earlier work is never summarized from scratch. The context meter shows how much of the window your next message will use, measured on what GDSense will actually send, and short notices tell you when a summary is coming or when a request needs trimming. Your full transcript always stays readable in History.
What does the context meter show?
Every model can consider a limited amount of text at once, called its context window. The meter under the prompt box shows how much of that window your next message will use: your draft, the recent conversation, the summary of older messages in a long chat, the open script and project map when they are due to be sent, and anything you attach.
The meter measures what GDSense will really send, not the full visible transcript. Once older messages are summarized, the meter drops to reflect the shorter request, and removing an attachment chip lowers it before you send.
- Below 80%, keep working normally.
- From 80%, a short line under the meter tells you when older messages will be summarized on your next send.
- Near 95%, GDSense lets you know the window is almost full and what will make the next request fit.
Keep in mind: The context meter describes the model’s context window, not your account. The separate credit bar shows your remaining balance.
How does GDSense summarize older messages?
When a conversation approaches the context window, your next send summarizes the older part of the conversation and keeps the latest few exchanges exactly as written. The answer then works from that summary plus the recent messages, so long sessions stay responsive and on topic.
The summary is a saved checkpoint, not a one-off. As the chat keeps growing, GDSense extends the existing summary with the messages that have since become older, instead of summarizing the whole conversation again. The summary is stored with the conversation in your local History, so loading a saved chat later picks up from the same point.
- 1
Keep chatting
Nothing needs to change in how you work. The hint under the meter tells you a summary is due.
- 2
Send your next message
GDSense summarizes older messages as part of that send and answers from the summary plus recent exchanges.
- 3
Continue later
Load the conversation from History and keep going; the saved summary comes with it.
What do the long-conversation notices mean?
Each notice describes one specific situation, so you always know whether to keep going, trim a request, or start fresh. When two apply to the same answer, they appear together in a single notice.
| Notice | What it means | What to do |
|---|---|---|
| Long Conversation | The conversation is using most of the context window and older messages will be summarized on your next send. | Nothing; keep working. |
| Context Window Almost Full | The next request nearly fills the window. | Send as usual in a long chat, or remove large attachments if a single request is very big. |
| Large Request | This one request is large and there are no older messages to summarize yet. | Shorten the message or remove an attachment with its ×. |
| Older Messages Left Out | To fit the window, some of the oldest input was not included for this answer. | Restate any early detail the answer needs, or start a new chat for a fresh topic. |
| Response Truncated | The answer reached its output limit and may end early. | Ask GDSense to continue, or narrow the request. |
| Conversation Too Long | The conversation has grown beyond what can be summarized. | Start a new chat; your message is held, not sent or charged, and History is unchanged. |
How do long Agent runs and promoted chats behave?
During an Agent run, the meter follows the run itself. A multi-step run summarizes its own older tool rounds as it works, so a long task keeps going without you trimming anything mid-run.
When you continue an earlier conversation in Agent Mode, or promote a Chat into Agent Mode, the conversation’s saved summary seeds the run along with the exchanges that came after it. The Agent starts with the same understanding the conversation had built up.
How can I keep a long conversation focused?
- Choose + New Chat when you switch to an unrelated task, so the new topic starts clean.
- Leave the open script and project map on; they are sent only when new or changed.
- Pin an important script with @openscript when later turns must keep that exact version in view.
- Restate a key requirement in a later message if the answer should honor it after many turns.
- Attach a large file only to the message that needs it, instead of repeating it in every follow-up.
Keep in mind: Summaries change what is sent, never what you see. The complete transcript remains readable and exportable from History.