DeepScientist

GitHub | 中文文档 | English Docs | Paper | Website

15-minute local setup · One repo per quest · Visible research progress · Human takeover anytime

Built-in runners: Codex, Claude Code, Kimi Code, OpenCode

Quick Start • Launch Your First Project • Product Tour • Codex Setup • Claude Setup • Kimi Setup • OpenCode Setup

Maintainers: BenchStore YAML Guide

May 12 update: v1.6.0 is available with Claude Code, OpenCode, Kimi Code, BenchStore, and science evidence workflows.

Unlike one-shot AI Scientist or autoresearch-style systems, DeepScientist is a local-first autonomous research studio that keeps the full loop moving on your machine, from baselines and experiment rounds to paper-ready outputs, with a 10-minute setup. Powered by Findings Memory, Bayesian optimization, and the Research Map, it keeps turning each new result into the next starting point and goes deep through broader exploration and, when needed, thousands of experiment validations.

If you want the technical deep dive behind DeepScientist, watch the Video.

---

https://github.com/user-attachments/assets/3c7abb44-2b25-4477-a011-10a3154d6d76

Still Spending Your Time On Research Grunt Work?

What drains researchers is often not the lack of ideas. It is the endless cycle of low-leverage work:

• new papers keep coming, but only a small fraction turns into an actionable next-step research plan
• baseline repos fail on environment, dependency, data, and script issues before real work even starts
• experiment results get scattered across terminals, scripts, notes, and chats, making later review painful
• writing, figures, and analysis live in separate tools, so turning them into a coherent paper takes far too long

This is the problem DeepScientist is built to solve:

turn fragmented, repetitive, easy-to-lose research work into a local AI workspace that can keep moving, keep accumulating, and keep getting stronger over time

DeepScientist Is Not Just Another "Research Chatbot"

It is not a tool that summarizes papers, throws you a few ideas, and leaves the dirty work to you.

It is much closer to a real long-running AI research partner:

About

DeepScientist is not a one-shot agent demo. It is a system built for long-horizon research work.

What Can It Actually Help You Get Done?

1. Start a real project from a paper or a research question

• feed it a core paper, a GitHub repository, or a natural-language research objective
• it turns those inputs into an executable quest instead of a chat that loses state after a few turns

2. Reproduce baselines and keep the reproduction reusable

• restore repositories, prepare environments, handle dependencies, and track the critical failures
• preserve what broke, what got fixed, and which steps are trustworthy for future rounds

3. Run experiments continuously instead of stopping after one pass

• propose the next hypothesis from existing results
• branch, ablate, compare, and record conclusions
• keep failed routes as assets instead of deleting them

4. Turn results into materials you can actually ship

• organize findings, conclusions, and analysis
• produce figures, reports, and paper drafts
• support local PDF and LaTeX compilation workflows

5. Follow the same research effort from multiple surfaces

• the web workspace in your browser
• the TUI workflow on a remote server
• external connector surfaces for collaboration and progress updates

The current docs already cover these collaboration channels:

• Weixin
• QQ
• Telegram
• WhatsApp
• Feishu
• Lingzhu / Rokid

Why Is It Easier To Keep Using?

What retains users is not a flashy demo. It is a system that becomes more useful the longer you work with it.

DeepScientist tends to stick for four reasons:

Local-first by default

• code, experiments, drafts, and project state stay on your own machine or server by default
• this is especially valuable for unpublished ideas, sensitive experiment history, and longer-running research loops

One repo per quest

• every quest is a real Git repository
• branches, worktrees, files, and artifacts naturally express research structure

The process is not a black box

• it does not only give you an output
• you can inspect what it read, what it changed, what it kept, and what it plans to do next

Human collaboration is built in

• DeepScientist can move autonomously
• you can also step in, edit, redirect, and hand control back whenever you want

Why Try It Now?

Because this is not just a concept. It is a real system with public docs, a public paper, and a public install path.

• 2026/03/24: DeepScientist officially released v1.5
• 2026/02/01: the paper went live on OpenReview for ICLR 2026
• npm install path is already available: @researai/deepscientist
• both Chinese and English docs are available, along with Web, TUI, and connector entry points

Product Preview

Architecture Overview

Example Outputs

Workspace Preview

Progress Reporting

Projects surface after long-running work

Who Will Love DeepScientist Most?

• graduate students and engineers who want to reproduce papers and push beyond existing baselines
• labs or research teams running long experiment loops, ablations, and structured result analysis
• people who want code, experiments, notes, and writing to live in one workspace
• users who do not want to hand unpublished ideas and intermediate results directly to a pure cloud workflow
• people who want to run work on servers while following progress from web, TUI, or messaging surfaces

The Core Philosophy Behind DeepScientist

We believe a system that is actually suitable for research should at least satisfy these principles:

• one quest, one repository, instead of letting everything dissolve after a short conversation
• branches and worktrees should express research routes naturally instead of being forced into chat history
• failed paths should be preserved, summarized, and reused instead of overwritten
• human researchers should always retain takeover power instead of being locked outside the loop
• the research process should be reviewable, inspectable, and auditable instead of relying on "the model says it did it"

If that sounds like the way you want to work, DeepScientist is worth trying now.

🚀 Get Started In 30 Seconds

If you want to try it right now, choose one of these two paths: run the npm commands yourself, or ask the coding tool you already use to install it for you.

Platform note: DeepScientist fully supports Linux and macOS. Native Windows support is currently experimental (strongly recommend WSL2).

Option 1: Manual Install With npm

Use this path when you already know which runner you want and prefer to control the install, login, and launch commands yourself.

DeepScientist ships four built-in runners:

• codex: use this when codex already works directly on your machine
• claude: use this when claude already works directly on your machine
• kimi: use this when kimi already works directly on your machine
• opencode: use this when opencode already works directly on your machine

If one of these CLIs already works for you, DeepScientist can usually meet you there instead of asking you to rebuild your whole setup first.

Think of the startup choice like this: bring one runner that already works, and DeepScientist gives you a persistent local research workspace around it.

If you just want the safest recommendation, start with Codex first.

🎯 Recommended first run: codex

npm install -g @researai/deepscientist
codex login
ds --here

If Claude Code already works directly in your shell, use this lane:

npm install -g @researai/deepscientist
claude --version
ds doctor --runner claude
ds --here --runner claude

If Kimi Code already works directly in your shell, use this lane:

npm install -g @researai/deepscientist
kimi --version
ds doctor --runner kimi
ds --here --runner kimi

If OpenCode already works directly in your shell, use this lane:

npm install -g @researai/deepscientist
opencode --version
ds doctor --runner opencode
ds --here --runner opencode

If you want to connect Gemini or Ollama, first use the runner-specific docs instead of guessing DeepScientist fields:

• Gemini: prefer OpenCode Setup
• Ollama: choose Codex, Claude Code, or OpenCode with Local Model Backends Guide

To stop the managed local daemon and all currently running agents:

ds --stop

🛠 Prefer installing from a Git checkout instead of npm? Use the repo path directly:

git clone https://github.com/ResearAI/DeepScientist.git
cd DeepScientist
bash install.sh
ds

Option 2: Let A Coding Tool Install It

Use this path when you already work inside Codex, Claude Code, OpenCode, Cursor, or another coding agent. There are only two steps:

1. Launch the coding tool in a directory where you are comfortable installing DeepScientist.
2. Copy and send this prompt:

Please install and launch DeepScientist on this machine. The official repo is https://github.com/ResearAI/DeepScientist and the docs start at https://github.com/ResearAI/DeepScientist/blob/main/docs/en/README.md . First inspect Node.js/npm, git, Python, OS, and shell environment. If global npm install is appropriate, run npm install -g @researai/deepscientist and verify ds --help. If source install is safer, git clone https://github.com/ResearAI/DeepScientist.git, cd DeepScientist, read the README, and run bash install.sh. After installation, confirm at least one runner works locally, such as codex, claude, opencode, or kimi; authenticate that CLI first, then run ds doctor --runner <name>, start with ds --here, and report the local URL plus the exact config docs I should read next.

If you plan to edit the UI or TUI from source, also install the workspace dependencies:

npm --prefix src/ui install
npm --prefix src/tui install

If you prefer the interactive first-run flow, run this once first:

codex

If codex still appears to be missing after installing DeepScientist, take the explicit repair path instead of assuming the bundled dependency was linked correctly:

npm install -g @openai/codex
which codex
codex login

If which codex still prints nothing after that, fix the npm global bin path first, then retry codex login and ds doctor.

Important runner note:

• DeepScientist can fall back to npm-bundled helper copies for codex, claude, and opencode when they are installed with the package. Kimi Code is treated as an external CLI unless a compatible local kimi helper is present.
• Runner authentication and provider configuration still belong to the underlying CLI. Make codex, claude, kimi, or opencode work once in your shell, then run ds doctor --runner <name>.
• You can also start DeepScientist first with the default runner and switch/configure Claude Code, Kimi Code, or OpenCode later from the web workspace settings.

After startup, the default local address is:

http://127.0.0.1:20999

Local browser auth is now optional and disabled by default. If you want a per-launch local access password, start with:

ds --auth true

Then you only need to do three things:

1. click Start Research
2. fill in the research goal, baseline links, paper links, or local paths
3. let DeepScientist start a real research project that can keep evolving locally

If this is your first run, prefer an isolated environment, a non-root user, and a local machine. For the full details, see:

• 00 Quick Start
• 15 Codex Provider Setup
• 24 Claude Code Setup
• 27 Kimi Code Setup
• 25 OpenCode Setup
• 09 Doctor

🧭 Choose Your Starting Path

⚡ I just want to get it running first

• 00 Quick Start
• 12 Guided Workflow Tour

🧪 I want to launch a real project today

• 02 Start Research Guide
• 01 Settings Reference

🖥 I mainly work on servers and terminals

• 05 TUI Guide Includes ds --tui --debug, redacted debug JSONL, and Web/TUI comparison guidance.

🔌 I want to connect my own models or external collaboration channels

• 15 Codex Provider Setup
• 24 Claude Code Setup
• 27 Kimi Code Setup
• 25 OpenCode Setup
• 21 Local Model Backends Guide
• Weixin Connector Guide
• QQ Connector Guide
• Telegram Connector Guide
• WhatsApp Connector Guide
• Feishu Connector Guide

🧠 I want to understand the system design first

• Docs Index
• Core Architecture Guide
• Prompt, Skills, and MCP Guide

Autonomous Research Systems

End-to-End Autonomous Research Systems

Documentation

• English Docs Index
• Chinese Docs Index

NLPCC 2026 AISB Challenge

If you want to benchmark or extend AI scientist systems in the wild, the NLPCC 2026 AISB shared task is a natural next stop:

• Registration
• Task Repository

For Developers And Maintainers

If you are developing or maintaining DeepScientist, continue with:

• Architecture
• Development Guide
• BenchStore YAML Guide
• CONTRIBUTING

Citation

If DeepScientist materially helps your paper, report, or research workflow, please cite the DeepScientist paper and disclose meaningful AI assistance honestly.

This is a strong request for fair academic attribution, not an extra software license condition.

Useful links:

• Paper: https://openreview.net/forum?id=cZFgsLq8Gs
• Repository citation metadata: CITATION.cff
• Citation and attribution guidance: docs/en/26_CITATION_AND_ATTRIBUTION.md
• Acknowledgements, including optional FermiLink science-workflow attribution: docs/en/99_ACKNOWLEDGEMENTS.md
• Name and logo usage: TRADEMARK.md

Suggested acknowledgment text:

We used DeepScientist to assist parts of the research workflow, including selected planning, implementation, experiment orchestration, analysis, and/or writing support. Final judgments, claims, and reported real experimental results remain the responsibility of the human authors.

DeepScientist is jointly developed by Yixuan Weng, Weixu Zhao, Shichen Li, Zhen Lin, and Minjun Zhu.

@inproceedings{
weng2026deepscientist,
title={DeepScientist: Advancing Frontier-Pushing Scientific Findings Progressively},
author={Yixuan Weng and Minjun Zhu and Qiujie Xie and QiYao Sun and Zhen Lin and Sifan Liu and Yue Zhang},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=cZFgsLq8Gs}
}

If this feels like the research workflow you have been waiting for, give the project a star. Every star makes it easier for more researchers who actually need it to find it.

Community

Welcome to join the WeChat group for discussion.

More From ResearAI

If you like DeepScientist, you may also want to explore the rest of the ResearAI ecosystem:

Roadmap

We are building DeepScientist as a long-term local-first research operating system.

The next major upgrades focus on four directions:

1. Deeper Research Loops

• AI Scientist Benchmark support for more realistic evaluation and comparison
• smoother automatic baseline upload, download, and reuse
• stronger experiment replay, comparison, and paper-facing outputs

2. Stronger Long-Horizon Memory

• stronger Memory and Findings Memory mechanisms
• better cross-run and cross-quest reuse
• less repeated failure and less rediscovery cost over long projects

3. Richer Multimodal And Collaborative Workflows

• VideoAnything-style multimodal research capabilities
• better local-model, connector, and copilot/autonomous collaboration flows
• a more efficient and more reliable DeepScientist system across local, collaborative, and long-horizon research settings

4. Stronger Security And Safer Deployment

• safer local-first and server-side deployment defaults
• stronger auth, permission, and connector-surface protection
• less fabrication, lower hallucination, and more verification-grounded outputs
• better auditability for long-running autonomous research workflows

If this direction is interesting to you, please give the project a Watch and a Star:

---

This project is maintained by WestlakeNLP. If you run into problems, please ask on DeepWiki first; if it still cannot be resolved, open an issue.

WestlakeNLP is led by ACL Fellow Professor Yue Zhang. If you are interested in a long-term internship, PhD position, or research assistant opportunity, contact Professor Yue Zhang at zhangyue@westlake.edu.cn.