AI-Researcher
https://github.com/hkuds/ai-researcher- Category
- AI Agents
- Rank
- No. 996Tools index
Previous survey · No. 991 ·
- Pricing
- Open Source
- Type
- AGENT
- Builder
- hkuds
- GitHub
- 5.7k stars
- Date
About
HKUDS's autonomous scientific innovation agent — proposes research questions, runs experiments, and writes up findings end-to-end.
What it does
AI-Researcher takes either a detailed machine-learning idea or a set of reference papers. It searches related repositories, downloads paper sources, chooses codebases and datasets, then coordinates specialist agents for planning, implementation, evaluation, analysis, and LaTeX manuscript assembly.
Why it's ranked here
The scope is unusually broad and backed by runnable packaging, a browser interface, container tooling, benchmark data, and distinct research and writing pipelines. That ambition comes with serious operational concerns: the browser interface installs hardcoded proxy settings, while the container command server accepts unauthenticated shell commands.
What's good
The two input modes support both researcher-led ideas and reference-led exploration. Separate agents handle preparation, surveying, planning, coding, judging, and experiment analysis. The manuscript pipeline composes six academic sections, cleans generated LaTeX, extracts bibliography entries, and compiles a PDF with references.
Tradeoffs
Setup is substantial. It requires Python 3.11 or newer, Playwright, API credentials, many dependencies, and a Docker environment configured for Linux on AMD64. Research tasks depend on predefined benchmark categories and instance data. The container command channel exposes shell execution without authentication, so network isolation is essential.
How to use it well
Use it for controlled machine-learning research prototypes where reference papers, benchmark definitions, container access, and human verification are available. Start with a detailed idea when requirements matter, or references when exploring directions. Keep it inside an isolated environment. It does not cover a hosted, low-setup research workflow.
Technical notes+
setup.cfg defines a setuptools package for Python >=3.11, three console entry points, and a large dependency set including LiteLLM, Gradio, Playwright, BrowserGym, ChromaDB, Docling, and sentence-transformers. main_ai_researcher.py dispatches detailed-idea, reference-led, and paper-generation modes through environment-driven configuration and a shared flag in global_state.py. research_agent/run_infer_idea.py and research_agent/run_infer_plan.py assemble cached tool and agent modules around repository search, paper-source download, dataset metadata, Docker execution, browsing, implementation, judging, and analysis. paper_agent/writing.py composes six sections before cleanup and compilation through paper_agent/writing_fix.py and paper_agent/tex_writer.py. web_ai_researcher.py sets fixed HTTP and HTTPS proxy addresses at import time. docker/tcp_server.py binds to 0.0.0.0, accepts raw commands, and passes them to Bash without authentication or command filtering.
Observed
- License
- MIT
- Primary language
- Python
- Python requirement
- Python 3.11 or newer
- Packaging
- Setuptools package with editable installation documented through uv
- Command-line interfaces
- Three console commands: ai-researcher, paper-agent, and benchmark
- Graphical interface
- Gradio web interface
- Container support
- Published Docker image or local Dockerfile build
- Configured platform
- Linux on AMD64
Read from README.md, setup.cfg, pyproject.toml, global_state.py, web_ai_researcher.py, main_ai_researcher.py, docker/tcp_server.py, paper_agent/writing.py, paper_agent/tex_writer.py, paper_agent/writing_fix.py, research_agent/constant.py, paper_agent/tex_writer_ori.py, paper_agent/section_composer.py, research_agent/run_infer_idea.py, research_agent/run_infer_plan.py.
What it can do
Generate research questions
Research domain or topic → Specific research questions
Design experiments
Research questions and objectives → Experimental methodology and protocols
Execute experiments
Experimental protocols and parameters → Raw experimental data and results
Analyze experimental data
Raw experimental data → Statistical analysis and insights
Write research papers
Experimental results and analysis → Formatted research manuscript
Conduct literature review
Research topic or keywords → Comprehensive literature analysis
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.