YggResearAI/DeepScientist
↗ GitHubNow, Stronger AI Pushes Frontiers, Stronger Our Shared Future.
1,294
Stars
114
Forks
8
Watchers
12
Open Issues
Safety Rating A
No hardcoded secrets, malicious code patterns, obfuscated logic, or prompt injection attempts were detected in the provided repository content. The project is a legitimate open-source research automation platform with a published ICLR 2026 paper, active documentation, and an npm package. The only minor note is the system's design gives AI agents broad filesystem and terminal access by default, which is inherent to its purpose rather than a security flaw. Overall the repository presents no red flags.
ℹAI-assisted review, not a professional security audit.
AI Analysis
DeepScientist is a local-first, autonomous AI research studio designed to support end-to-end scientific research workflows. It orchestrates long-running AI agents across the full research lifecycle — from ingesting papers and reproducing baselines to running iterative experiments, tracking results, and producing paper drafts. Each research project ("quest") is backed by a real Git repository with branches and worktrees for structured exploration. The system exposes a web workspace UI, a TUI for remote servers, and connectors to messaging platforms (Telegram, WhatsApp, WeChat, Feishu, QQ). It depends on OpenAI Codex CLI as its agent execution backend and is installed via npm.
Use Cases
- Reproducing academic paper baselines and tracking dependency/environment failures for future reuse
- Running iterative, multi-hypothesis ML experiments with branching, ablation, and result comparison
- Generating paper drafts, figures, and LaTeX/PDF outputs from accumulated experimental findings
- Monitoring long-running research workflows remotely via messaging platforms (Telegram, WhatsApp, WeChat)
- Organizing research history into a persistent, inspectable Git-based workspace with durable state
Tags
Security Findings (1)
The project depends on @openai/codex as its agent execution backend, installed via npm. No specific CVEs are visible from the README alone, but the dependency on a rapidly-evolving CLI tool with broad system access (file writes, terminal execution) represents a supply-chain risk worth monitoring.
Project Connections
karpathy/autoresearch
Both systems automate iterative AI-driven research loops (modify → run → evaluate → repeat), but DeepScientist provides a much richer end-to-end platform with persistent state, a web UI, messaging connectors, and paper writing support, while autoresearch is a minimal single-GPU experiment loop.
K-Dense-AI/claude-scientific-skills
Claude Scientific Skills provides 136 domain-specific science skill modules (bioinformatics, cheminformatics, materials science, etc.) that could be loaded into DeepScientist's agent workflows to expand its capability across specialized scientific domains.
PaperDebugger/paperdebugger
PaperDebugger handles in-editor academic writing assistance and review workflows in Overleaf, while DeepScientist handles experiment execution and result accumulation. The two tools address adjacent stages of the research-to-publication pipeline.
HKUDS/LightRAG
LightRAG's knowledge graph RAG pipeline could serve as a retrieval backend for DeepScientist's literature intake and baseline reproduction phases, enabling richer semantic search over accumulated papers and experimental notes.
uditgoenka/autoresearch
DeepScientist's autonomous iterative experiment loop (propose hypothesis → run → evaluate → branch) is conceptually inspired by Karpathy's autoresearch pattern, which uditgoenka's Claude Code skill also implements. DeepScientist generalizes this into a full research studio.