> For the complete documentation index, see [llms.txt](https://docs.cherryai.com.cn/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.cherryai.com.cn/docs/en-us/advanced-basic/chat/context-queue.md).

# Long Conversations, Context, and Queued Messages

The longer the conversation, the more history the model needs to read. When context usage is close to the limit, earlier content may no longer be able to participate in the answer. Rather than repeatedly appending “continue,” it is better to periodically organize the conclusions and unresolved issues.

<figure><img src="/files/0448a345943a8e022b65316b708432bcc5f6f9cf" alt="助手高级设置中的模型温度和上下文管理选项"><figcaption><p>Adjust the advanced settings only when you need to change the answer style or the way long conversations are handled; if unsure, keep the current values.</p></figcaption></figure>

### Managing long conversations

{% stepper %}
{% step %}

#### 1. Watch the context indicator

When the interface indicates rising context pressure, stop adding large attachments first and check which parts of the history are still relevant to the current goal.
{% endstep %}

{% step %}

#### 2. Have the model generate a handoff summary

Ask it to separately list “confirmed facts, current conclusions, unresolved issues, and constraints that must not be lost.” This is better suited for continuing work than a normal “summarize it.”
{% endstep %}

{% step %}

#### 3. Start a new topic to continue

Put the handoff summary and necessary files into a new topic, and state in the first message which goal will be handled next. Keep the original topic as a reference.
{% endstep %}
{% endstepper %}

### Adjust global context management

Path: \[Settings] → \[General] → \[Context Management]. These settings apply to regular assistant conversations; individual assistants can also override them in their own settings.

| Configuration item                | Product default          | Recommended starting point     | Function                                                            | Applicable scenario                                      | Notes                                                                          |
| --------------------------------- | ------------------------ | ------------------------------ | ------------------------------------------------------------------- | -------------------------------------------------------- | ------------------------------------------------------------------------------ |
| Number of recent messages to keep | Unlimited                | No limit for now               | Only send the most recent few messages to the model                 | Short tasks with a fixed window                          | Too small a value will make the model forget requirements that are still valid |
| Enable context management         | Enable                   | Keep it on                     | Manage very long tool results and history compression               | Tasks with long conversations and many tool calls        | If turned off, large results are more likely to fill up the context            |
| Tool output truncation threshold  | 50,000 characters        | Keep default                   | After large results are stored, let the model read them in segments | Web pages, logs, and long document tool results          | Does not mean the original result is deleted                                   |
| Automatic compression             | Enable                   | Keep it on                     | Summarize earlier history when approaching the window limit         | Continuous multi-turn work                               | Compression is a summary and does not guarantee every detail is preserved      |
| Compression model                 | Follow the current model | Follow the current model first | Choose the model used to generate the history summary               | Needed when you want separate control over speed or cost | Switching models increases debugging variables                                 |

\[Settings] → \[Default Model] also provides \[Model Call Retry]. It is off by default; when enabled, it will try up to 3 times by default using exponential backoff, and you can also choose fallback models in sequence. Retry and fallback only take effect before the model starts outputting; they will not switch a half-generated answer to another model to continue.

### Use the message queue

While the model is still replying, you can add the next request to the queue. This is suitable for appending one clear follow-up action, such as “After finishing, organize it into three conclusions.” If an Agent is running and you want to immediately correct the current direction, use the \[Guidance Shortcut]; if you need to start over completely, stop the current generation and restate the goal.

<figure><img src="/files/2d7e082c4cba72f769a6c049855aeac33c3d77ec" alt="消息队列中两条待发送消息和恢复自动发送按钮"><figcaption><p>When paused, you can first check the result; after resuming, queued messages will continue to be sent from top to bottom.</p></figcaption></figure>

In the figure: ① two queued messages in the current topic; ② automatic sending resumes. After resuming, messages will continue to be sent from top to bottom.

{% hint style="info" %}
The queue is not an automation plan. It only handles follow-up messages in the current topic; if you need to run at a fixed time, use \[Scheduled Tasks].
{% endhint %}

#### Use case: reviewing a long report

First upload the report and ask for issues to be listed by section. While the model is processing, add “organize the risk list after completion” and “generate a checklist at the end” to the queue in order. If you need to check the first round of results first, you can pause automatic sending; after confirming everything is correct, resume. This way, you do not need to stay in the conversation and send messages one by one, and you can also avoid the next request being executed before the review is complete.

<details>

<summary>Why did the model suddenly forget the requirements it had mentioned earlier?</summary>

First check whether the conversation is too long, whether the model was switched, and whether the key requirements appeared only once. Write stable rules into the explicit instructions for the current task; if you need long-term reuse, put them into an Agent prompt or skill.

</details>

<details>

<summary>Why did the model suddenly forget the requirements it had mentioned earlier?</summary>

First check whether the conversation is too long, whether the model was switched, and whether the key requirements appeared only once. Write stable rules into the explicit instructions for the current task; if you need long-term reuse, put them into an Agent prompt or skill.

</details>


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.cherryai.com.cn/docs/en-us/advanced-basic/chat/context-queue.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.