# Chat memory and summarizing

> What the chat memory ring means, why long chats cost more, and how summarizing frees memory and saves quota.

## Chat memory is not your usage limit

The ring beside the message box is **chat memory**: how much of this chat RoGenie can keep in mind at once, measured in tokens. Hover or tap it to see the numbers, for example "Chat memory 40% full: 80K of 200K tokens."

It is **not** your quota. Your quota is the amount of model usage your plan allows per 5 hours and per month; see [How usage is counted](https://rogenie.ai/docs/usage-and-quota). Chat memory is about one conversation.

Each model has its own memory size (its *context window*). You can also set a lower ceiling yourself in [Agent settings](https://rogenie.ai/docs/agent-settings#context-and-savings).

## Why long chats cost more

Every step RoGenie takes sends the conversation so far to the model. The longer the chat, the more tokens each step uses, so a very long chat spends your quota faster for the same work. Summarizing keeps that in check.

## Summarizing

When a chat's memory fills up, RoGenie **summarizes older messages automatically**: it replaces the older turns with a compact summary and keeps the recent ones in full, so the conversation can keep going. A *Summarization* step appears in the work log while it happens, and "Chat summarized" when it is done.

You can also summarize yourself at any time:

- type `/summarize` in the message box, or
- open the chat's actions menu and choose **Summarize to free memory**.

When a chat gets large, RoGenie may suggest it: "This chat's memory holds N tokens. Summarizing now compacts the older turns, which slows how fast your quota is used and keeps the model sharp." You can summarize from the suggestion or dismiss it.

Good to know:

- Summarizing makes one model call, which uses a little quota.
- Nothing important is paraphrased away: under the summary, the card keeps **Pinned facts** (details RoGenie pinned during the chat so they survive summarizing) and **Your requests, verbatim**, and both go to the model with the summary.
- Your history stays in the chat: the summarized turns show read-only beside a summary card, and they stay [searchable](https://rogenie.ai/docs/chats-and-search). Only what the model is sent changes.
- If summarizing is slow, RoGenie says "Summarization is taking longer than expected. It may still finish — reload to check." If it fails, the chat is left as it was and nothing is lost.

## When a chat no longer fits

If a chat has grown beyond what the selected model can hold, the run fails with a message that the conversation no longer fits the model's context window. Then:

1. Type `/summarize` to compact the chat, and try again, or
2. switch to a model with a larger memory, or
3. start a new chat. Lasting facts about your place belong in [project notes](https://rogenie.ai/docs/memory#project-notes), which every new chat reads.

## Starting fresh

A new chat starts with empty chat memory but still knows your [standing instructions, your notes and the place's project notes](https://rogenie.ai/docs/memory). For a new feature or a new topic, a new chat is often cheaper and sharper than continuing a very long one.