Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/IconDean/research-agent/llms.txt

Use this file to discover all available pages before exploring further.

The Deep Research Agent ships with a full-stack web interface: a FastAPI backend that streams Server-Sent Events (SSE) and a React frontend built with Vite. Together they give you a live, browser-based research experience — submit a question, watch the agent plan and search in real time, then read the rendered markdown report without leaving your browser.

Starting the Application

You need two terminal windows running concurrently — one for the API server and one for the frontend dev server.
1

Start the FastAPI backend

Open a terminal, navigate into the research_agent directory, and start Uvicorn:
cd research_agent
uvicorn server:app --reload --host 127.0.0.1 --port 8000
Uvicorn will reload automatically when server-side files change. The API listens on http://127.0.0.1:8000.
2

Start the React frontend

Open a second terminal, navigate into the frontend directory, and start Vite:
cd frontend
npm run dev
Vite serves the UI at http://localhost:5173 and proxies all /api/* requests to the backend automatically.
3

Open the UI in your browser

Navigate to http://localhost:5173. If the API health indicator in the top-right corner shows green, the frontend has successfully reached the backend and you are ready to run research.
In development mode, Vite proxies every request to /api/* directly to http://127.0.0.1:8000. This means the frontend and backend can run on different ports without CORS issues during development — you never need to change the API base URL in your code.

Submitting a Research Question

The main input area is a textarea in the center of the page. Type your question, then start the research in one of three ways:
  • Type freely — enter any research question in the textarea.
  • Use an example chip — click one of the three pre-set example questions to populate the textarea instantly:
    • “What are the latest breakthroughs in fusion energy?”
    • “How is CRISPR being used to treat genetic diseases in 2025?”
    • “What are the main approaches to scalable quantum error correction?”
  • Click Start — press the Start button below the textarea to begin.
Press Ctrl+Enter (or Cmd+Enter on macOS) while the textarea is focused to submit your question immediately — no need to reach for the mouse.

Reading the Activity Log

Once a research run begins, a live activity log appears below the input area. Each line represents a progress event streamed from the backend via SSE, tagged with an event type so you can follow exactly what the agent is doing.
Event typeWhat it meansDisplay format
startResearch has begunResearch started
planAgent has created a research planPlan: <detail>
searchAgent is issuing a web searchSearching: <detail>
fetchAgent is reading a webpageReading: <detail>
thinkingAgent is reasoning or evaluating results<detail> (shown directly)
gapsAgent identified gaps and is planning follow-upGaps: <detail>
blockA URL was rejected by the credibility filterBlocked: <detail>
Events appear in real time as the agent works through its research iterations, giving you a transparent view of every search query issued, every page read, and every credibility decision made.

Cancelling a Research Run

If you want to stop an in-progress research run — for example because you phrased the question incorrectly — click the Cancel button that replaces the Start button during an active run. The frontend immediately aborts the in-flight SSE connection via an AbortController, clears the loading state, and leaves whatever activity log entries were already received in place. You can then edit your question and start a new run straight away.

Reading the Rendered Report

When the agent finishes, the final report appears below the activity log as rendered markdown — headings, bullet points, bold text, tables, and inline links are all formatted by react-markdown with remark-gfm. You do not need to switch to a separate view or download a file; the report renders directly on the page. Inline citations and hyperlinks in the report are clickable and open in a new tab, letting you verify sources without leaving the interface.

API Health Indicator

A small status indicator in the UI checks the backend’s /api/health endpoint every 15 seconds. When the backend is reachable the indicator shows green; if the backend goes offline or returns an error it turns red. If you see a red indicator, confirm that Uvicorn is still running in Terminal 1 before submitting a question.

Build docs developers (and LLMs) love