Free plan chat memory fills after ten messages

Seeing "chat memory full" after only a handful of exchanges points to a hard limit on the free tier. The user reports that after roughly ten messages the interface flashes the error, even though they are on a free subscription. Earlier they had a conversation that exceeded one thousand messages, which triggered the same message. They tried to work around it by pasting a summary of that long thread into a new chat, but the error appeared again after about ten messages, and they repeated the attempt four or five times with the same result.
The pattern suggests that the free plan enforces a strict token ceiling on the context window. Once the accumulated input—both the conversation history and any newly added text—passes that ceiling, the system refuses further input and surfaces the “chat memory full” notice. Because the summary itself still consumes tokens, inserting it does not reset the window enough to allow many more turns before the limit is hit again.
A practical way to stay under the ceiling is to ask the model to compress the ongoing dialogue into a concise summary before the window fills. The following prompt instructs the model to retain only the essential points while discarding redundant phrasing, thereby shrinking the token count:

You are a helpful assistant. Summarize the conversation so far in no more than three sentences. Focus on the main topics, any decisions made, and open questions. Do not add new information.

When this prompt is sent, the model returns a short paragraph that captures the gist of the exchange. Because the summary is much shorter than the raw dialogue, feeding it back as the seed for a fresh chat dramatically reduces the token load. Users have observed that after inserting such a summary they can continue for another round of exchanges before the warning reappears, effectively extending the usable length of a free‑tier session.
The approach works for two reasons. First, the explicit length constraint (“no more than three sentences”) forces the model to trim filler. Second, by directing the model to focus on decisions and open questions, the summary preserves the information needed to resume the discussion without re‑explaining already settled points. This keeps the context window lean enough to stay within the free plan’s limits while still allowing meaningful progress.
If the summary still proves too large, one can further tighten the request—for example, asking for a single‑sentence recap or specifying bullet points. Each iteration reduces the token footprint, giving the user more room to continue chatting before encountering the “chat memory full” error again. This method does not require upgrading the plan; it simply leverages the model’s own summarization ability to manage the constrained context window on the free tier.