[NeurIPS2025] "AI-Researcher: Autonomous Scientific Innovation" -- A production-ready version: https://novix.science/chat
5.6k
Stars
706
Forks
67
Open issues
4
Contributors
AI Analysis
AI-Researcher is a system for autonomous scientific research automation that accepts research ideas or reference papers and orchestrates end-to-end workflows from concept through publication, using AI agents to generate novel research strategies and implementations. It is purpose-built for academic researchers and AI labs seeking to accelerate research cycles; it is not a general-purpose coding tool or infrastructure platform.
Inferred from signals mentioned in the README (tests, CI, type safety) — not a review of the actual code.
AI's overall editorial judgment — not an average of the bars above, can weigh other factors too.
HKUDS AI-Researcher automates the full scientific research pipeline from idea to manuscript
AI-Researcher is a Python-based agentic system that automates end-to-end scientific research: literature review, idea generation, algorithm design, implementation, validation, and paper writing. It targets academic researchers and AI labs seeking to accelerate experimentation cycles. Accepted as a NeurIPS 2025 Spotlight paper, it has a companion hosted product at novix.science. With ~5,500 stars and 703 forks since March 2025, it has attracted meaningful academic and hobbyist interest, though verifiable production deployment at scale is limited beyond the hosted demo.
Launched March 2025 by HKUDS (Hong Kong University Data Science lab), the project evolved from an initial release to a comprehensive upgrade in May 2025 with an arXiv paper, benchmark suite, and web GUI, before earning NeurIPS 2025 Spotlight status in September 2025.
Initial traction came from the HKUDS lab's prior reputation (their AI-Trader repo has 20K+ stars). The NeurIPS 2025 Spotlight acceptance in September 2025 likely provided a credibility boost. Growth appears steady but not explosive — 38 stars in the last 7 days suggests the initial viral spike has settled into a slower, sustained interest curve typical of academic tools.
A hosted production version exists at novix.science/chat, indicating at least one real deployment pathway. The NeurIPS 2025 paper provides academic validation. However, verifiable evidence of broad production usage by external organizations or teams beyond the authors is not documented in the README. Adoption not verified at scale.
Likely a multi-agent orchestration pipeline built in Python, with distinct modules per research phase (literature review, ideation, coding, validation, manuscript generation). Based on README, it appears to accept two input modes: detailed idea descriptions and reference-paper-based ideation. Integration with external LLM APIs is likely required. The hosted product at novix.science suggests a web frontend exists separately from the open-source core.
Not documented in README
Last push was October 16, 2025 — approximately 8.5 months before the current date (July 2026). This represents a meaningful gap with no recent commits visible in metadata. The project may be in a maintenance-only or wind-down phase, or development may have shifted primarily to the commercial novix.science product. Community channels (Slack, Discord, WeChat, Feishu) were established, but their current activity level cannot be verified from available metadata.
ADOPT IF: you are an academic researcher or AI lab wanting to automate repetitive research pipeline steps (literature review, prototyping, paper drafting) and can tolerate dependency on external LLM APIs and a codebase with no recent commits since October 2025. AVOID IF: you need active maintenance guarantees, enterprise-grade support, or a tool verified in large-scale production environments — the ~8.5-month commit gap and migration of energy toward the commercial novix.science product create sustainability uncertainty. MONITOR IF: the novix.science product open-sources more of its stack or the repo resumes active development; a NeurIPS Spotlight paper means the underlying research is solid enough to revisit.
Independent dimensions
Mainstream potential
4/10
Technical importance
7/10
Adoption evidence
3/10
- Last code push was October 2025 (~8.5 months ago), raising concern that open-source development has stalled or shifted to the closed commercial product at novix.science.
- End-to-end research automation pipelines are highly sensitive to upstream LLM API changes; without active maintenance, compatibility may degrade over time.
- Reproducibility and reliability of AI-generated scientific manuscripts and experiments are inherently difficult to validate, posing risks if outputs are used uncritically.
- The academic niche limits the contributor base, making it harder to sustain community-driven development compared to general-purpose developer tools.
- No license information is documented, creating legal ambiguity for organizations that want to deploy or build on the codebase commercially.
Likely to remain a respected academic reference implementation cited alongside the NeurIPS paper, while active product development continues through the commercial novix.science platform rather than the open-source repo.
Newsletter
Get analyses like this every Monday
Free weekly digest of the most interesting open-source discoveries.
Languages
Information
- Website
- https://arxiv.org/abs/2505.18705
- Language
- Python
- Last updated
- 9mo ago
- Created
- 16mo ago
- Analyzed with
- anthropic/claude-haiku-4-5
Stars over time
Contributors over time
Top 100 contributors only — repos with more will plateau at 100.
Open issues
Acid geopolmer
khalilkraft
OpenAIBox — LLM dimension analysis tool, potential integration with AI-Researcher
License clarification + interest in building a multimodal extension (with a contribution back)
No API found
Top contributors
Recent releases
No releases published yet.
Similar repos
HKUDS/Auto-Deep-Research
Auto-Deep-Research is an open-source AI research assistant that automates deep...
ResearAI/DeepScientist
DeepScientist is a local-first autonomous research studio that automates the...
SakanaAI/AI-Scientist-v2
AI Scientist-v2 is an autonomous research system that generates scientific...
| Repository | Stars | Week Δ | Language | Score | Updated |
|---|---|---|---|---|---|
|
HKUDS/AI-Researcher
this
|
5.6k | +55 | Python | 7/10 | 9mo ago |
|
HKUDS/Auto-Deep-Research
|
1.6k | — | Python | 7/10 | 9mo ago |
 OpenRaiser/NanoResearch
|
1.5k | — | Python | 7/10 | 2mo ago |
|
ResearAI/DeepScientist
|
3.2k | — | TypeScript | 7/10 | 2w ago |
|
SakanaAI/AI-Scientist-v2
|
6.8k | — | Python | 7/10 | 7mo ago |
|
HKUDS/AI-Trader
|
20.7k | — | Python | 7/10 | 4w ago |
AI-Scientist-v2 has ~6,700 stars and similar end-to-end research automation goals, backed by Sakana AI as a company. AI-Researcher differentiates with its two-level input model and NeurIPS peer review, but AI-Scientist benefits from higher commercial backing and slightly higher star count.
With 90K+ stars, autoresearch dominates mindshare in this space. AI-Researcher is far behind in adoption but appears more comprehensive in its pipeline scope (including manuscript writing), suggesting a different depth-vs-breadth tradeoff.
deep-research (19K stars, TypeScript) focuses primarily on literature synthesis rather than full pipeline automation. AI-Researcher covers more of the research lifecycle but may be harder to integrate into non-Python workflows.
Smaller project (3.1K stars, TypeScript) targeting similar automation goals. AI-Researcher has stronger academic provenance and a published benchmark, giving it more credibility for research-oriented users.
From the same lab, AI-Trader has 20K+ stars, suggesting the HKUDS team knows how to build popular tools. AI-Researcher's narrower academic audience likely explains its comparatively modest star count, not necessarily lower quality.
