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

# Deep Research

> Run multi-step agentic research that produces comprehensive, cited reports on complex topics.

Deep Research is Hadiq.io's agentic research mode. Instead of returning a single LLM response, it runs a structured multi-step process: it clarifies the question, builds a research plan, dispatches parallel research agents, and synthesizes all findings into a detailed cited report.

<Note>
  Hadiq.io Deep Research ranked at the top of the [deep research leaderboard](https://github.com/Hadiq.io-dot-app/Hadiq.io_deep_research_bench) as of February 2026.
</Note>

## When to use Deep Research

| Use Deep Research                                | Use regular chat                      |
| ------------------------------------------------ | ------------------------------------- |
| Complex, multi-part questions                    | Simple factual questions              |
| Topics that span multiple sources                | Questions with a clear, direct answer |
| Research that benefits from structured synthesis | Quick back-and-forth conversation     |
| Reports you need to share or reference later     | Exploratory or iterative queries      |

Deep Research takes longer than a standard chat response — typically several minutes — because it runs multiple research cycles. Use it when depth and completeness matter more than speed.

## How it works

<Steps>
  <Step title="Clarification (optional)">
    Before researching, Deep Research may ask a clarifying question if your query is ambiguous. Answer it to help the system focus its research on what you actually need.
  </Step>

  <Step title="Research plan">
    The system generates a research plan — a structured outline of the topics and sub-questions it will investigate. You can see this plan as it is generated in the UI.
  </Step>

  <Step title="Parallel research cycles">
    Research agents run in parallel, each investigating a portion of the plan. Agents can use internal search (your connected knowledge), web search, and URL fetching. Multiple cycles may run, with the orchestrator reassessing after each round and dispatching follow-up research as needed.
  </Step>

  <Step title="Final report">
    Once the research cycles are complete, the system synthesizes all findings into a structured report with citations. Every claim is traceable back to its source.
  </Step>
</Steps>

## Trigger Deep Research

<Tabs>
  <Tab title="From chat">
    In any chat session, click the **Deep Research** button in the toolbar (or select it from the mode picker) before sending your message. Your query will enter the Deep Research flow instead of returning a standard chat response.
  </Tab>

  <Tab title="From an agent">
    If an agent has web search or internal search tools enabled, you can trigger Deep Research from within the agent's chat. Select **Deep Research** from the toolbar and send your query.
  </Tab>
</Tabs>

<Tip>
  Frame your Deep Research query as a research question or report brief for the best results. For example: *"Summarize the current state of our Q1 deal pipeline, including top risks and opportunities, based on our Salesforce data and recent call notes."*
</Tip>

## What sources Deep Research can use

Deep Research only uses search tools — it does not execute code or call external action tools. The sources available during a research session depend on what is enabled:

* **Internal search** — searches your connected and indexed knowledge sources
* **Web search** — fetches live information from the web
* **URL fetching** — opens and reads specific URLs discovered during research

To make your internal knowledge available in Deep Research, ensure the relevant connectors are set up and indexed.

[Set up connectors →](/features/connectors)

## Requirements

Deep Research requires an LLM with at least **50,000 input tokens** of context. Models with smaller context windows cannot run Deep Research. If you see an error about context limits, ask your admin to configure a model with a larger context window.