Open Deep Research

An AI-powered research assistant that performs iterative, deep research on any topic by combining search engines, web scraping, and large language models.

The goal of this repo is to provide the simplest implementation of a deep research agent - e.g. an agent that can refine its research direction over time and deep dive into a topic. Goal is to keep the repo size at <500 LoC so it is easy to understand and build on top of.

If you like this project, please consider starring it and giving me a follow on X/Twitter. This project is sponsored by Aomni.

How It Works

flowchart TB
    subgraph Input
        Q[User Query]
        B[Breadth Parameter]
        D[Depth Parameter]
    end

    DR[Deep Research] -->
    SQ[SERP Queries] -->
    PR[Process Results]

    subgraph Results[Results]
        direction TB
        NL((Learnings))
        ND((Directions))
    end

    PR --> NL
    PR --> ND

    DP{depth > 0?}

    RD["Next Direction:
    - Prior Goals
    - New Questions
    - Learnings"]

    MR[Markdown Report]

    %% Main Flow
    Q & B & D --> DR

    %% Results to Decision
    NL & ND --> DP

    %% Circular Flow
    DP -->|Yes| RD
    RD -->|New Context| DR

    %% Final Output
    DP -->|No| MR

    %% Styling
    classDef input fill:#7bed9f,stroke:#2ed573,color:black
    classDef process fill:#70a1ff,stroke:#1e90ff,color:black
    classDef recursive fill:#ffa502,stroke:#ff7f50,color:black
    classDef output fill:#ff4757,stroke:#ff6b81,color:black
    classDef results fill:#a8e6cf,stroke:#3b7a57,color:black

    class Q,B,D input
    class DR,SQ,PR process
    class DP,RD recursive
    class MR output
    class NL,ND results

Loading

Features

• Iterative Research: Performs deep research by iteratively generating search queries, processing results, and diving deeper based on findings
• Intelligent Query Generation: Uses LLMs to generate targeted search queries based on research goals and previous findings
• Depth & Breadth Control: Configurable parameters to control how wide (breadth) and deep (depth) the research goes
• Smart Follow-up: Generates follow-up questions to better understand research needs
• Comprehensive Reports: Produces detailed markdown reports with findings and sources
• Persistent Reports: Completed reports are saved to disk and can be downloaded via /api/reports/{id}
• Concurrent Processing: Handles multiple searches and result processing in parallel for efficiency

Requirements

• Node.js environment
• API keys for:
  • Firecrawl API (for web search and content extraction)
  • OpenAI API (for o3 mini model)

Setup

Node.js

1. Clone the repository
2. Install dependencies:

npm install

3. Set up environment variables in a .env.local file:

FIRECRAWL_KEY="your_firecrawl_key"
# If you want to use your self-hosted Firecrawl, add the following below:
# FIRECRAWL_BASE_URL="http://localhost:3002"

OPENAI_KEY="your_openai_key"
ACCESS_KEY="choose_a_strong_key"
# Optional: control how many learnings are processed per report section
# REPORT_CHUNK_SIZE="20"

To use local LLM, comment out OPENAI_KEY and instead uncomment OPENAI_ENDPOINT and OPENAI_MODEL:

• Set OPENAI_ENDPOINT to the address of your local server (eg."http://localhost:1234/v1")
• Set OPENAI_MODEL to the name of the model loaded in your local server.

Docker

1. Clone the repository

2. Rename .env.example to .env.local and set your API keys

3. Run docker build -f Dockerfile

4. Run the Docker image:

docker compose up -d

5. Execute npm run docker in the docker service:

docker exec -it deep-research npm run docker

Usage

API Server (Default)

Start the API server:

npm start

The server will run on port 3051. See the Web API section below for endpoints.

CLI Mode

For interactive command-line research:

npm run cli

You'll be prompted to:

1. Enter your research query
2. Specify research breadth (recommended: 3-10, default: 4)
3. Specify research depth (recommended: 1-5, default: 2)
4. Answer follow-up questions to refine the research direction

The assistant asks clarifying questions before starting the actual research. Once you provide your answers, it begins gathering data and compiling the report.

The system will then:

1. Generate and execute search queries
2. Process and analyze search results
3. Recursively explore deeper based on findings
4. Generate a comprehensive markdown report

The final report will be saved as report.md or answer.md in your working directory, depending on which modes you selected.

Concurrency

If you have a paid version of Firecrawl or a local version, feel free to increase the ConcurrencyLimit by setting the CONCURRENCY_LIMIT environment variable so it runs faster.

If you have a free version, you may sometimes run into rate limit errors, you can reduce the limit to 1 (but it will run a lot slower).

DeepSeek R1

Deep research performs great on R1! We use Fireworks as the main provider for the R1 model. To use R1, simply set a Fireworks API key:

FIREWORKS_KEY="api_key"

The system will automatically switch over to use R1 instead of o3-mini when the key is detected.

Custom endpoints and models

There are 2 other optional env vars that lets you tweak the endpoint (for other OpenAI compatible APIs like OpenRouter or Gemini) as well as the model string.

OPENAI_ENDPOINT="custom_endpoint"
CUSTOM_MODEL="custom_model"

Web API

The API server starts automatically with npm start on port 3051 (or the value of the PORT environment variable).

All requests must include an Authorization header containing the value of your ACCESS_KEY.

Start a report job

Send a POST request to /api/jobs with the research query and optional breadth and depth:

curl -X POST http://localhost:3051/api/jobs \
  -H 'Content-Type: application/json' \
  -d '{"query":"Quantum computing trends","breadth":3,"depth":2}'

The response contains a jobId and a list of follow-up questions that must be answered before the research begins.

Submit your answers to /api/jobs/{jobId}/answers:

curl -X POST http://localhost:3051/api/jobs/<jobId>/answers \
  -H 'Content-Type: application/json' \
  -d '{"answers":["first answer","second answer"]}'

Once the answers are submitted, the job status changes to pending and the system starts researching.

Poll for results

Fetch /api/jobs/{jobId} until the status changes to completed:

curl http://localhost:3051/api/jobs/<jobId>

When finished, the response will include a reportUrl field. Fetch this URL to download the saved Markdown report:

curl http://localhost:3051/api/reports/<jobId> > report.md

How It Works

1. Initial Setup

   • Takes user query and research parameters (breadth & depth)
   • Generates follow-up questions to understand research needs better

2. Deep Research Process

   • Generates multiple SERP queries based on research goals
   • Processes search results to extract key learnings
   • Generates follow-up research directions

3. Recursive Exploration

   • If depth > 0, takes new research directions and continues exploration
   • Each iteration builds on previous learnings
   • Maintains context of research goals and findings

4. Report Generation

   • Compiles all findings into a comprehensive markdown report
   • Includes all sources and references
   • Organizes information in a clear, readable format

License

MIT License - feel free to use and modify as needed.