> 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/agent-workspace/permissions-memory-background.md).

# Permissions, Memory, and Background Tasks

Agents can perform file, terminal, and network operations, and can also call sub-agents and background tasks. Permissions determine whether you need to ask first, memory determines whether it will still know next time, and the status panel on the right tells you what it is doing now.

<figure><img src="/files/315801e21ae2b11b8e84a047ded5c04ff060761e" alt="Agent 可选的仅规划、默认、自动审批和完全访问权限模式"><figcaption><p>Daily use starts with default permissions; plan-only is for when you want to review the approach first, and full access is only for tasks with clear boundaries and recoverable changes.</p></figcaption></figure>

### Permission mode

| Mode                 | Behavior                                           | Applicable scenario                                      | Notes                                                                                            |
| -------------------- | -------------------------------------------------- | -------------------------------------------------------- | ------------------------------------------------------------------------------------------------ |
| \[Ask each time]     | Ask before editing files or running commands       | Default starting point, real project directory           | When there are many operations, confirmations will be frequent, but it is the easiest to control |
| \[Auto-accept edits] | Can edit files, ask before running commands        | Document cleanup, reversible code changes                | Check the working directory and version control status first                                     |
| \[Smart approval]    | Whether to allow it is determined by safety checks | Supported by the model and the task boundaries are clear | Some models may still ask each time                                                              |
| \[Plan only]         | Plan only, do not edit files                       | Plan review, pre-launch review                           | Suitable for reviewing the plan before executing it                                              |
| \[Full Access]       | Skip permission checks                             | An isolated, trusted, and recoverable environment        | May delete files or access the network; highest risk                                             |

The permission scope provided by different run modes is not exactly the same: Pi does not offer \[Plan only]; new Pi Agents use \[Smart approval] by default; DeepSeek Harness does not offer \[Smart approval]; \[Enhanced: Claude Agent] provides the full set of five modes.

Configuration path: left navigation \[Work] → Agent menu → \[Edit] → \[Basic] → \[Permission mode].

{% hint style="danger" %}
Channels can override the Agent's permission mode separately. External chat entry points are more likely to receive unexpected instructions, so you should usually choose \[Inherit agent settings] or a mode stricter than the Agent's; do not open full access just to click fewer confirmations.
{% endhint %}

### Agent memory

Agent memory follows the Agent, not a specific task or working directory. It is suitable for storing long-term preferences, project facts, technical decisions, and experience; one-off progress is recorded as time-stamped logs for retrieval in later tasks.

Tell the Agent directly in \[Work]:

```
Remember: all Chinese copy released externally should use full-width punctuation, and exaggerated headlines should not be used. Future related tasks should follow this rule.
```

When correction is needed, clearly state that the old information is no longer valid and ask to update the memory. Do not write passwords, API keys, private identity information, or short-term irrelevant content into long-term memory.

### Sub-agents, workflows, and background commands

For complex tasks, the Agent can delegate research, organization, and verification to sub-agents, or orchestrate multiple steps through workflows. Long-running commands can run in the background without blocking the entire conversation. After you confirm, the Agent can also start another conversation or send a task to an existing conversation; the request returns immediately, and the result comes back to the initiating conversation later, preserving the source and delivery status.

In the right-side \[Status], you can view:

* Tasks in progress and completed;
* Sub-agents and workflows;
* Background commands and stop entry points;
* Success, failure, and count of tool calls;
* Context usage and declared artifacts.

After turning on \[Settings] → \[Notifications] → \[Conversation completion notifications], when you switch to another tab or window to work, you will receive a system notification when the assistant reply is complete, the Agent task is complete, or approval is pending; clicking the notification will take you back to the corresponding conversation.

<figure><img src="/files/cde6077f074e8ee59ce67e0bd26d30a8aa55bfa5" alt="Agent 状态面板中的子任务、后台命令和上下文用量"><figcaption><p>Long tasks can view artifacts, subtasks, background commands, and context usage in [Status].</p></figcaption></figure>

### User case: continuously maintaining project standards

The team writes stable coding conventions into Agent memory and turns detailed review steps into a skill. Each time they review a new branch, they open a task and let the Agent use sub-agents to check interfaces, data migrations, and tests separately, then the main Agent summarizes the conclusions. When the rules change, update the memory; there is no need to modify the historical records of every task.

<details>

<summary>Can memory replace a knowledge base?</summary>

No. Memory is suitable for a small amount of stable facts and experience that will be used across tasks; a knowledge base is suitable for structured documentation and can control the Agent's retrieval scope.

</details>

<details>

<summary>The task seems stuck. Where should I look first?</summary>

First open \[Status] on the right and check whether it is waiting for permission, whether a background command is still running, whether a subtask failed, or whether the provider is retrying. If you need deeper request details, then open developer mode to view the call chain.

</details>

<details>

<summary>The task seems stuck. Where should I look first?</summary>

First open \[Status] on the right and check whether it is waiting for permission, whether a background command is still running, whether a subtask failed, or whether the provider is retrying. If you need deeper request details, then open developer mode to view the call chain.

</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/agent-workspace/permissions-memory-background.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.