> 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/knowledge-base/agent.md).

# Use with Agent

After binding a knowledge base to an Agent, the Agent can search and read authorized materials in multi-step tasks; when it is clear that the materials need to be updated, knowledge base management can also be enabled.

{% hint style="info" %}
Ordinary conversations are suitable for one-time Q\&A, while Agents are suitable for ongoing research, comparison, file generation, and step-by-step acceptance. An Agent only accesses the knowledge bases bound in its configuration.
{% endhint %}

## The difference between ordinary conversations and Agents

| Ordinary conversation                                      | Agent                                                         |
| ---------------------------------------------------------- | ------------------------------------------------------------- |
| Temporarily select a knowledge base before each message    | Long-term binding in the Agent configuration                  |
| Suitable for immediate Q\&A and short comparisons          | Suitable for multi-step research and file delivery            |
| Mainly answers using retrieved excerpts                    | Can search materials and manage knowledge bases by permission |
| The current conversation determines the scope of materials | The Agent configuration determines the accessible scope       |

## Configure a read-only knowledge base Agent

{% stepper %}
{% step %}

### 1. First make sure the knowledge base has passed acceptance

Confirm that the target materials are ready, and use real questions to complete retrieval testing. The Agent cannot make up for missing full text or incorrect chunking.
{% endstep %}

{% step %}

### 2. Open the Agent editing window

Go to \[Work], select the target Agent, and open \[Edit Agent] from the menu.
{% endstep %}

{% step %}

### 3. Bind the minimum necessary knowledge base

Open the \[Knowledge Base] tab and click \[Add Knowledge Base]. Only bind the libraries needed for this task to avoid interference between materials from different departments or versions.
{% endstep %}

{% step %}

### 4. Enable knowledge base search

Open \[Built-in Tools] and enable \[Knowledge Base Search]. For read-only research, Q\&A, summarization, and comparison, this is usually enough.

<figure><img src="/files/a01523c20a15851f6f4d60c5871d8a843eccc272" alt="Agent 内置工具设置中的知识库搜索和知识库管理能力"><figcaption><p>Search is for reading, management is for changing materials; start with the minimum permissions by default.</p></figcaption></figure>
{% endstep %}

{% step %}

### 5. Test with a clearly bounded task

Ask the Agent to list sources first, then give the conclusion; if the materials do not support it, state that clearly and do not fill in the gaps from common sense.
{% endstep %}

{% step %}

### 6. Verify sources and deliverables

Check which knowledge base each conclusion comes from, whether facts and recommendations in the file are separated, and whether unsupported parts are marked.
{% endstep %}
{% endstepper %}

## How to choose between built-in tools, knowledge bases, skills, and MCP

<figure><img src="/files/9b00652317bad3dcfaeb99d65a0f86a7f0cab964" alt="说明 Agent 内置工具、知识库、技能和 MCP 各自用途的选择关系图"><figcaption><p>Use a knowledge base to look up information, skills for repeatable methods, and MCP to access external systems; do not use expanded permissions to replace a clearly defined task.</p></figcaption></figure>

## Knowledge base search and management

| Capability                | What it can do                                   | Applicable tasks                       | Default recommendation                            |
| ------------------------- | ------------------------------------------------ | -------------------------------------- | ------------------------------------------------- |
| Knowledge base search     | Search, list, and read bound knowledge bases     | Q\&A, research, summaries, comparisons | Keep enabled                                      |
| Knowledge base management | Add, delete, or refresh knowledge base documents | Approved material maintenance          | Disabled by default, enable temporarily as needed |

{% hint style="warning" %}
Binding a knowledge base only grants access scope; it does not copy out another knowledge base. After materials are updated or re-indexed, the Agent's next search will use the updated content.
{% endhint %}

{% hint style="danger" %}
After enabling \[Knowledge Base Management], adding, deleting, and refreshing will change the materials or the index. Before approval, confirm the target knowledge base, the specific entries, how same-name conflicts will be handled, and the rollback plan.
{% endhint %}

## How the Agent's configurations work together

<figure><img src="/files/292b5ee745cdde007a6d0f2e66377958134a094a" alt="模型分工、可用能力和安全边界共同作用于 Agent 任务与交付结果的架构图"><figcaption><p>The model determines understanding and generation, the knowledge base provides evidence, and permissions determine how far the Agent can act.</p></figcaption></figure>

## Recommended task templates

### Research and generate a report

> Find all rules about overseas business trip approvals and insurance in the bound knowledge bases. First list the sources and conflict points, then generate a Markdown checklist. Do not fill in unsupported content.

### Update FAQ

> Search the existing items for accommodation reimbursement and compare the latest policy with the old FAQ. First provide a list of proposed changes; after approval, refresh the relevant documents.

### Comparison across multiple knowledge bases

> Find evidence separately from the "Product Manual" and "After-sales Cases" knowledge bases, and organize it into three columns: "Official rules / Real cases / Suggested wording." Keep the source name for each conclusion.

## Configuration notes

| Configuration item        | Recommended starting point                      | When to add                                       | Risk control                                                |
| ------------------------- | ----------------------------------------------- | ------------------------------------------------- | ----------------------------------------------------------- |
| Bind knowledge base       | 1 task-related library                          | There is a real need for cross-library comparison | Write the purpose of each library clearly in the prompt     |
| Knowledge base search     | Enable                                          | As long as the task requires looking up materials | Verify whether the sources come from the bound scope        |
| Knowledge base management | Disable                                         | Explicitly need to add, delete, or refresh        | Approve item by item, and back up important materials first |
| Output requirements       | Separate facts, inferences, and recommendations | Need to generate a report or file                 | Keep the source name for each fact                          |

## User case

Xiaolin bound \[Official Manual] and \[Review Cases] to the after-sales Agent, and only enabled knowledge base search. He asked the Agent to list safety warnings, official steps, and case recommendations by device model, and keep the three separate. Only when he found old cases that needed updating did he temporarily enable the management tool, first review the list of proposed changes, and then approve it.

The acceptance standard is: the Agent does not access unbound materials, does not write case recommendations as official rules, and all write operations have a clear target and acceptance result.

## Result acceptance

* Sources come only from the knowledge bases currently bound to the Agent.
* Search results cover every condition in the task.
* The deliverable separates material facts, Agent inferences, and recommendations.
* The management operation explains the target, impact, and result.
* After updating the materials, rerun the fixed retrieval questions.

## Common questions

<details>

<summary>Why can ordinary conversations find it, but the Agent cannot?</summary>

Check whether the target knowledge base is bound to the current Agent, and whether \[Knowledge Base Search] is enabled. The bound scope of different Agents is not inherited from one another.

</details>

<details>

<summary>When should knowledge base management not be enabled?</summary>

For read-only research, team-shared policy libraries, and material libraries that retain historical versions, only search is enabled by default. When updates are needed, temporarily enable management and approve item by item.

</details>

<details>

<summary>What should I do if the Agent wants to operate on an unbound knowledge base?</summary>

Do not expand to all knowledge bases. After confirming the task truly requires it, add the target library to the current Agent, or switch to an Agent that already has that library bound.

</details>

## Continue reading

<table data-view="cards"><thead><tr><th></th><th></th><th data-hidden data-card-target data-type="content-ref"></th></tr></thead><tbody><tr><td><strong>Use in conversation</strong></td><td>Complete one immediate, source-backed Q&#x26;A.</td><td><a href="/pages/b797f738b8c03492018a83b57869ba2f43a00a62">/pages/b797f738b8c03492018a83b57869ba2f43a00a62</a></td></tr><tr><td><strong>Data, privacy, and maintenance</strong></td><td>Understand the boundaries of permissions, cloud services, and backups.</td><td><a href="/pages/79d7d1d4aa8046a5b156be2497b3f18795dd88a8">/pages/79d7d1d4aa8046a5b156be2497b3f18795dd88a8</a></td></tr><tr><td><strong>Knowledge base application cases</strong></td><td>Refer to after-sales and research workflows.</td><td><a href="/pages/f5585710b91e24d4206ea9a29da14da13f79dbe1">/pages/f5585710b91e24d4206ea9a29da14da13f79dbe1</a></td></tr></tbody></table>


---

# 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/knowledge-base/agent.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.