> For the complete documentation index, see [llms.txt](https://docs.maiagent.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.maiagent.ai/maiagent-user-guide/en/work/maigpt/deep-research.md).

# Deep Research

Create a research plan in MaiGPT, review estimated Credits and output formats, then read the research report

Deep Research is MaiGPT's in-depth research feature. Enter a research topic, and AI proposes a research plan. It then searches for and organizes information based on the direction you confirm and produces a report. It is suitable for market research, competitive analysis, and technical research.

## How It Differs from Regular Chat <a href="#comparison-with-regular-chat" id="comparison-with-regular-chat"></a>

|                  | Regular Chat                            | Deep Research                                                                  |
| ---------------- | --------------------------------------- | ------------------------------------------------------------------------------ |
| Best suited for  | Quick questions and everyday tasks      | Research that requires searching and organizing multiple sources               |
| How it works     | Enter a question and receive a response | Review the research plan and cost estimate first, then choose an output format |
| Research results | Chat response                           | Canvas document, chat response, or downloadable files                          |
| Waiting time     | Usually shorter                         | Varies with the topic's scope and search conditions                            |

## When to Use It <a href="#when-to-use" id="when-to-use"></a>

A corporate planning manager is preparing to evaluate how to adopt generative AI. He enables <mark style="color:blue;">Deep Research</mark> in MaiGPT and enters "Research the current state, challenges, and success stories of generative AI adoption among Taiwanese businesses." He first reviews the research plan and estimated Credits, then selects <mark style="color:blue;">Reply directly in the chat</mark>. Once the research is complete, he can read the organized findings and cited sources in the original conversation as a basis for further discussion.

***

## Start a Research Task <a href="#start-a-research" id="start-a-research"></a>

The following example covers topic research without attached files.

### 1. Enable Deep Research <a href="#click-deep-research-button" id="click-deep-research-button"></a>

Open a <mark style="color:blue;">New Chat</mark> in <mark style="color:blue;">MaiGPT</mark>, click <mark style="color:blue;">Add</mark> (＋) next to the input box, and select <mark style="color:blue;">Deep Research</mark>. Once enabled, a label with the same name appears below the input box.

<figure><img src="https://1360999650-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6v6TNkkOQVfRYfcNirHL%2Fuploads%2Fgit-blob-b4a5b4186a5735a0703c0669655cf5941a207a2b%2Fdeep-research-entry-prod.png?alt=media" alt="Deep Research entry in the Add menu"><figcaption><p>Select "Deep Research" in the Add menu</p></figcaption></figure>

{% hint style="info" %}
If you cannot see <mark style="color:blue;">Deep Research</mark>, contact your organization administrator to check your [MaiGPT role permissions](/maiagent-user-guide/en/org/roles/maigpt-role-permissions.md).
{% endhint %}

### 2. Enter a Research Topic <a href="#enter-research-topic" id="enter-research-topic"></a>

Describe the topic, scope, and purpose of the report, then click the send button at the bottom right of the input box. If you want to confirm the direction first, explicitly request, "Please propose a research plan first and wait for my confirmation before proceeding."

| More general          | More specific                                                                                                      |
| --------------------- | ------------------------------------------------------------------------------------------------------------------ |
| "Research AI"         | "Research the current state, challenges, and success stories of generative AI adoption among Taiwanese businesses" |
| "Analyze competitors" | "Compare the enterprise knowledge base features of three specified products and summarize their use cases"         |

<figure><img src="https://1360999650-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6v6TNkkOQVfRYfcNirHL%2Fuploads%2Fgit-blob-07dfa96b10f6a93388c1fe3938e3660855ccb702%2Fdeep-research-input-prod.png?alt=media" alt="Input box with Deep Research enabled and a research topic entered"><figcaption><p>Confirm that "Deep Research" is enabled before submitting your research topic</p></figcaption></figure>

### 3. Review the Plan and Cost Estimate <a href="#review-research-plan" id="review-research-plan"></a>

Read the research direction proposed by AI and the estimated Credits. If the scope is unsuitable, request adjustments in the same conversation first. Proceed to the next step once both the direction and the estimate are acceptable.

<figure><img src="https://1360999650-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6v6TNkkOQVfRYfcNirHL%2Fuploads%2Fgit-blob-28d9408e878f9d401a71f655e9c0fdfd99c044e6%2Fdeep-research-plan-estimate-prod.png?alt=media" alt="List of research directions and estimated Credits usage"><figcaption><p>The research plan includes a cost estimate; the amount shown is only an estimate for this demonstration</p></figcaption></figure>

### 4. Choose an Output Format and Confirm the Start <a href="#choose-output-format" id="choose-output-format"></a>

Select a format in the <mark style="color:blue;">Choose the research report's output format before confirming the start</mark> card:

| Option                                                      | How results appear after research is complete                                          |
| ----------------------------------------------------------- | -------------------------------------------------------------------------------------- |
| <mark style="color:blue;">Canvas document</mark>            | Preview the report in Canvas, view its source, read it in full screen, or download it  |
| <mark style="color:blue;">Reply directly in the chat</mark> | Display the report in the original conversation for convenient reading within the chat |
| <mark style="color:blue;">Download files (docx／PDF)</mark>  | Generate downloadable Word and PDF files for saving or sharing                         |

<figure><img src="https://1360999650-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6v6TNkkOQVfRYfcNirHL%2Fuploads%2Fgit-blob-cb6056fc0b61ec7f285956ff99da97fc9647af5a%2Fdeep-research-output-format-prod.png?alt=media" alt="Card offering three research report output formats"><figcaption><p>Choose Canvas, a chat response, or downloadable files based on the report's purpose</p></figcaption></figure>

{% hint style="warning" %}
**Selecting a format also sends a confirmation message.** The system then checks the available quota and starts the research if the check passes. Review the plan and cost estimate before selecting a format. Once selected, the card retains your chosen format, and you cannot select another format on the card.
{% endhint %}

<figure><img src="https://1360999650-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6v6TNkkOQVfRYfcNirHL%2Fuploads%2Fgit-blob-b6d07b8fd0357eac5a994054c5357a655b7020c3%2Fdeep-research-format-selected-prod.png?alt=media" alt="Card with Reply directly in the chat selected"><figcaption><p>After selection, the card retains only the output format used for this research</p></figcaption></figure>

### 5. Wait for the Research Results <a href="#wait-for-ai-completion" id="wait-for-ai-completion"></a>

Once research starts, the conversation displays <mark style="color:blue;">Research progress</mark>. AI searches for and organizes information, then generates a report. The time required varies with the research scope, sources, and generation conditions. Refer to the actual progress shown.

***

## Credits Usage Estimates and Quota Prechecks <a href="#credit-estimation" id="credit-estimation"></a>

Deep Research may perform multiple searches and analyses. The cost estimate in the plan helps you decide whether to proceed. When you confirm execution, the system checks the available quota against the estimate.

### How the Estimate Is Calculated <a href="#how-estimation-works" id="how-estimation-works"></a>

* The estimate first uses the median actual usage from up to 20 of your organization's most recent research tasks.
* If your organization has no research history, it uses the median of recent research tasks across the platform. If there is still no history, it uses the default estimate configured in the system.
* This is a reference based on historical usage, rather than a fixed price quoted for the current topic. Actual charges depend on usage such as searches and Tokens, and may be higher or lower than the estimate.

{% hint style="warning" %}
**The estimate is not a spending limit and does not guarantee that you will have sufficient quota throughout the research.** The check before confirmation uses the estimate and balance at that time. Subsequent actual usage may still exceed the estimate.
{% endhint %}

### The Research Plan Includes a Cost Estimate <a href="#estimate-in-plan" id="estimate-in-plan"></a>

When AI proposes a plan, it includes estimated Credits and a note that actual usage may be higher or lower than the estimate. AI generates this text, so the wording and amount may vary. Read the notice in your current conversation.

### Quota Check Before Execution <a href="#quota-precheck" id="quota-precheck"></a>

After you confirm execution, the system checks the organization wallet and the quotas applicable to this task. If any applicable quota is insufficient, the research will not start:

| Check                       | What is checked                                                               | Details                                                                                           |
| --------------------------- | ----------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------- |
| Organization wallet         | Whether available Credits can cover the estimated usage                       | [About Credits](/maiagent-user-guide/en/org/credits.md)                                           |
| Individual quota            | The member's available quota, if configured                                   | [Member Credits quotas](/maiagent-user-guide/en/org/credits/member-credit-quota.md)               |
| Shared role quota           | The available shared quota of the member's role, if configured                | [Role quotas](/maiagent-user-guide/en/org/credits/member-credit-quota.md#role-quota)              |
| Conversation platform quota | The available quota for the platform used by this conversation, if configured | [Conversation platform Credits quotas](/maiagent-user-guide/en/org/credits/inbox-credit-quota.md) |

### What Happens If the Check Blocks Research <a href="#blocked-experience" id="blocked-experience"></a>

* AI displays a notice about insufficient quota. Follow the notice and contact your organization administrator for assistance.
* Research does not start. The original conversation and research plan remain at the stage awaiting confirmation.
* Once the quota has been replenished, ask AI to proceed in the same conversation. For example, "The quota has been replenished. Please start the research according to the plan you just proposed." You do not need to enter the topic again.

### What to Do When Quota Is Insufficient <a href="#insufficient-quota-actions" id="insufficient-quota-actions"></a>

| Your role                  | What to do                                                                                                                                                                |
| -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Regular member             | Share the insufficient quota notice with your organization administrator and explain the research purpose and estimated usage                                             |
| Organization administrator | Follow the notice to check the organization wallet, individual quota, role quota, or conversation platform quota, then take action according to the organization's budget |

### Frequently Asked Questions <a href="#credit-estimation-faq" id="credit-estimation-faq"></a>

#### Q: Does seeing a cost estimate mean that amount has already been deducted? <a href="#faq-estimate-no-charge" id="faq-estimate-no-charge"></a>

The estimated amount is not deducted directly. **The estimate and quota check themselves are free, but AI responses such as generating the research plan and confirming details in the conversation may still consume Credits.** Therefore, research that has not started or is blocked by the quota check does not mean the entire conversation is free.

#### Q: Why is actual usage higher or lower than the estimate? <a href="#faq-estimate-vs-actual" id="faq-estimate-vs-actual"></a>

The estimate uses historical research costs. Differences in the number of searches, source lengths, and Token usage for this task lead to differences in actual usage.

#### Q: Do I need to start a new research task after replenishing the quota? <a href="#faq-resume-after-topup" id="faq-resume-after-topup"></a>

No. Return to the original conversation and ask AI to proceed according to the confirmed plan.

#### Q: Why do I not see a cost estimate? <a href="#faq-no-estimate-shown" id="faq-no-estimate-shown"></a>

Deployments without Credits billing enabled do not provide cost estimates. If an estimate is temporarily unavailable, the system still checks the existing available quota rules. The absence of an estimate does not mean the task is free.

***

## Read and Export the Report <a href="#read-and-export-report" id="read-and-export-report"></a>

### Read the Report <a href="#read-report" id="read-report"></a>

Once research is complete, open the results in the format you selected earlier. As you read the conclusions, also check the cited sources and the scope of applicability provided in the report.

<figure><img src="https://1360999650-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6v6TNkkOQVfRYfcNirHL%2Fuploads%2Fgit-blob-2041d9a320de86a843311e2dee943608fbe6c25d%2Fdeep-research-report-chat-prod.png?alt=media" alt="Beginning of a research report displayed directly in the chat"><figcaption><p>When a chat response is selected, the completed report can be read in the original conversation; the image shows an excerpt from the beginning</p></figcaption></figure>

When you select <mark style="color:blue;">Canvas document</mark>, the report appears in the Canvas panel. Click <mark style="color:blue;">Full screen view</mark> at the top right to enlarge it for reading. <mark style="color:blue;">Switch to code mode</mark> displays the Markdown source. Click <mark style="color:blue;">Switch to preview mode</mark> to return to the formatted view.

<figure><img src="https://1360999650-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6v6TNkkOQVfRYfcNirHL%2Fuploads%2Fgit-blob-11cbf4f3f94ea32b66f5f5e3909d3fd32d25bd4a%2Fdeep-research-canvas-preview-prod.png?alt=media" alt="Research report displayed in Canvas full screen preview"><figcaption><p>Canvas full screen preview preserves the report title, body, and action buttons at the top</p></figcaption></figure>

### Edit the Report <a href="#edit-report" id="edit-report"></a>

The Markdown Canvas used for research output on this page provides preview, source viewing, and download features. **It does not support editing by typing directly.** To edit the report manually, download the original Markdown file, or select <mark style="color:blue;">Download files (docx／PDF)</mark> before starting the research and edit the downloaded Word file with an appropriate editor.

For additional searches or analysis, ask follow-up questions in the conversation. New AI responses are still charged based on actual usage.

### Export <a href="#export" id="export"></a>

If you select <mark style="color:blue;">Download files (docx／PDF)</mark>, <mark style="color:blue;">Generating downloadable files…</mark> appears after the report is complete. Wait until the files have been generated, then click <mark style="color:blue;">Download</mark> for the corresponding file. If you select Canvas, click <mark style="color:blue;">Download</mark> at the top right of the panel and choose <mark style="color:blue;">Original format (.md)</mark> or <mark style="color:blue;">PDF document</mark>.

<figure><img src="https://1360999650-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6v6TNkkOQVfRYfcNirHL%2Fuploads%2Fgit-blob-d5982518ba62074eebf39c8dc9a3c34244a1dc9f%2Fdeep-research-file-downloads-prod.png?alt=media" alt="Word and PDF files available for download after research is complete"><figcaption><p>Once the files have been generated, click "Download" next to the Word or PDF file</p></figcaption></figure>

<figure><img src="https://1360999650-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F6v6TNkkOQVfRYfcNirHL%2Fuploads%2Fgit-blob-fb03d6723986b955c93a2b65a523306d3eff9944%2Fdeep-research-canvas-download-menu-prod.png?alt=media" alt="Canvas download menu for original Markdown and PDF"><figcaption><p>The Canvas download menu offers the original Markdown and a PDF document</p></figcaption></figure>

## Next Steps <a href="#next-steps" id="next-steps"></a>

* [Learn about Canvas](/maiagent-user-guide/en/work/maigpt/canvas.md)
* [View Credits and usage history](/maiagent-user-guide/en/org/credits.md)
* [Return to the MaiGPT overview](/maiagent-user-guide/en/work/maigpt.md)


---

# 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 by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.maiagent.ai/maiagent-user-guide/en/work/maigpt/deep-research.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

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.