Navigate Select ESC Close

Open Deep Research

2025-02-20 Science & Technology
46.7k
1.2k
46
LangChain
LangChain
193.0k subscribers

Unlock all features

FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.

Description

AI assistants capable of in-depth, autonomous ("deep") research on user-supplied topics has become a major area of interest. Here, we discus a few of the common architectures across various closed / open deep research tools and provide an overview of our implementation. We show how to run it locally using LangGraph studio. We also review many of the configurations, allowing users to customize the input, the models, the search API, the report structure, and the depth of research. Repo: https://github.com/langchain-ai/open_deep_research Video notes: https://mirror-feeling-d80.notion.site/Deep-Research-18f808527b17803b809af5e42f81f5fc?pvs=4 Chapters: 00:00 - Introduction to Deep Research 00:45 - Live Demo in LangGraph Studio 01:30 - Human Feedback Loop for Report Planning 02:15 - Parallel Deep Research Process 03:00 - Core Components of Deep Research Systems 04:00 - Comparing Human-in-Loop Approaches 05:00 - Tool-Calling Agents vs Workflows 06:15 - Popular Open Source Implementations 07:30 - Architectural Tradeoffs 08:00 - Evaluations (Gaia and Humanities Last Exam) 09:00 - Configurability Advantages 10:00 - Setting Up Custom Configurations 10:45 - Example Report Comparison 12:00 - Cost Analysis (vs $200/month subscriptions) 12:45 - Benefits of Open Source Research Tools

Top Comments (10)

@PlinioS.Borges 2025-02-20

Great work! I think if your tool have a more developed citation implementation, it would be great for academic researches like me. Thnak you

8
@MitkoPioneer 2025-02-20

Excellent summary and distinction! We've been building a bunch of similar workflows at Pioneer, and now we're looking for a product manager who likes to thinker with what makes a good workflow.

7 1 replies
@CookiesDarkMatter 2025-02-24

And just like that, you have cracked perplexity :P

5 1 replies
@MrSur512 2025-03-09

Wow!! Such a density of raw condensed information packed into 13 min. Appreciate it Lance

3
@jacobc7731 2025-02-20

This is gonna be great to explore. Great job!

3
@bimbotsolutions6665 2025-02-27

Your model consideration page was very informative ,please do this all the time

1
@BrunoAguirre 2025-02-24

very insightful! I'm about to try it out!

1
@snoopthebandit 2025-04-14

A way you guys can pull ahead of SoTA. Is creating a document library based Deep Research. Simply replacing this with the sources, using vector stores or graphing to research the library. Maybe potentially coupled with Data from sources too. Just an idea, I am too busy to work this out right now. And I wish I could within a short turnover time. But a team might be able to. In Europe this is extremely important as every company wants to work with internal documents and can't share these outside of their tenant. To solve this, they either host opensource models or use the OpenAI Azure Models (o3, o1, 4o) to ensure data compliancy. If a solution like this could be deployed it would mean lot's of usecases.

1
@danielf7235 2025-07-20

fantastic video. an absolutely complete dive into deep research. can’t get better than that.

1
@domenicofumagalli8781 2025-03-29

It’s pretty crazy to think that this could actually be treated as a component in langgraph ❤

0

Unlock the Data Inside
Turn Videos into Knowledge

  • Get FREE 10/day: transcripts, summaries, chats
  • Chat with videos, export text & PDF
  • $1 free API credit for RAG, chatbots & research

Free forever plan • All features unlocked

App screenshot