> ## Documentation Index
> Fetch the complete documentation index at: https://linkly.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Deep Research Assistant

> Let AI conduct topic-based research and analysis using real materials on your computer, instead of generating content from thin air.

## Let AI Truly Understand Your Materials

Your computer holds a wealth of valuable materials — industry reports, research papers, project documents, meeting minutes, competitive analyses... When you need to conduct a focused research task, these materials are the best source material.

But the problem is: AI can't see them.

You open ChatGPT or Claude, and it can only answer based on its training data. Want it to reference your materials? You'd have to manually open files, copy content, and paste it into the chat. When you have a dozen or even dozens of documents, this approach becomes practically impossible.

**Linkly AI changes this.**

## Pain Points of Traditional Approaches

<CardGroup cols={2}>
  <Card title="Manual copy and paste" icon="paste">
    Opening files one by one, copying key content, pasting into the AI chat.
    Inefficient and easy to miss important information.
  </Card>

  <Card title="Context window limitations" icon="window-maximize">
    AI has limited context — you can't fit all your materials at once. You can
    only select a few, but you're never sure if you've missed something
    critical.
  </Card>

  <Card title="AI can't proactively retrieve" icon="robot">
    Traditional AI can only passively receive what you give it. It can't
    proactively browse your materials or determine which ones are relevant.
  </Card>

  <Card title="Format barriers" icon="file-pdf">
    Many materials are in PDF or Word format. After direct upload, AI parsing
    quality varies — especially for scanned PDFs.
  </Card>
</CardGroup>

## Linkly AI's Solution

After installing Linkly AI and connecting its MCP service to your AI assistant, AI gains the ability to **proactively search your local documents**.

Linkly AI provides three progressive tools that let AI work like a skilled research assistant:

<Steps>
  <Step title="search — Find relevant documents">
    Based on the research topic, AI uses keyword or semantic search to find
    relevant files in your local documents. Like an assistant going to the file
    room to browse the catalog and compile a list of potentially useful files.
  </Step>

  <Step title="outline — View document outlines">
    For documents found in the search, AI can first view the outline (titles,
    chapter structure) to quickly determine if they're worth reading in depth.
    Like an assistant scanning the table of contents to decide which chapters
    need careful review.
  </Step>

  <Step title="read — Read specific content">
    Once confirmed as relevant, AI reads the specific content of the file and
    extracts the needed information. Like an assistant carefully reading and
    taking notes.
  </Step>
</Steps>

This progressive approach is both efficient and precise — AI doesn't read all files at once (that would be too slow and wasteful). Instead, it strategically goes deeper step by step. This also avoids the fragmentation issues and high costs of traditional RAG retrieval.

## Practical Example

### Scenario: Summarize Q1 Progress Based on Project Documents

Suppose you're a project manager with various project documents from the past quarter on your computer — weekly reports, meeting minutes, milestone reports, etc. You need to write a Q1 summary.

In your local AI assistant (Claude Code, ChatGPT Codex, Cursor, etc.), type:

```
Based on the project documents on my computer, please summarize the project progress for 2025 Q1.
Focus on: milestones completed, issues encountered, and next steps.
Produce a summary with source citations, use linkly-ai.
```

The AI Agent will automatically call Linkly AI's search tool, using keywords like "Q1," "milestone," and "progress" to search your documents.
After identifying relevant documents, it extracts specific content through the `outline` and `read` tools, assembles the complete context, and produces the summary.

## Advanced Usage

<Tip>These tips can make AI research even more effective.</Tip>

* **Define the research scope**: Specify in your prompt what aspects you want AI to focus on and what to ignore — this makes the search more precise
* **Specify file types**: If you know the materials are in PDF or Word format, tell the AI, and it will prioritize searching those types
* **Ask in steps**: For complex research tasks, break them into sub-questions and ask each one separately, focusing on one topic at a time
* **Request source citations**: Ask AI to note which file each piece of information came from, making it easy to verify

## Applicable Scenarios

<CardGroup cols={2}>
  <Card title="Competitive analysis" icon="chart-line">
    Have AI read your collected competitor reports and industry analyses to
    produce a structured competitive comparison
  </Card>

  <Card title="Investment research" icon="money-bill-trend-up">
    Based on multiple research reports and financial data, let AI help you
    outline investment logic and risk factors
  </Card>

  <Card title="Academic literature review" icon="graduation-cap">
    Have AI read multiple papers and summarize the current state of research,
    methodology comparisons, and research gaps
  </Card>

  <Card title="Project retrospective" icon="clipboard-check">
    Based on past project documents, weekly reports, and meeting minutes,
    automatically generate project summaries and retrospective reports
  </Card>
</CardGroup>

<Note>
  Linkly AI itself does not call any LLM services. It only indexes and retrieves
  your local documents — the actual "thinking" and "analysis" is performed by
  the AI assistant of your choice.
</Note>

<Note>
  Your original file content is never uploaded. The LLM only receives document
  fragments and cannot modify the originals or know their actual location,
  maximizing your data privacy and security. Additional privacy protection
  mechanisms, such as sensitive data masking, will be added in future updates.
</Note>