SpectrAI-Initiative/InnoClaw

What it solves

InnoClaw provides a unified, self-hostable workspace for scientific research. It eliminates the need to switch between disparate tools for literature review, note-taking, coding, and experiment execution by integrating them into a single AI-native environment grounded in the user's own local files.

How it works

The system maps server-side folders into persistent workspaces. It uses RAG (Retrieval-Augmented Generation) to allow users to chat with their documents with citations. It integrates literature search across sources like ArXiv and PubMed, and employs a multi-agent system for "Deep Research"—where specialized roles (e.g., moderator, skeptic) execute multi-phase research workflows. For execution, it can submit and monitor remote jobs via Shell, Slurm, or Kubernetes.

Who it’s for

• Researchers needing grounded AI assistance for paper study and literature review.
• ML/AI Engineers requiring a workspace for code and data management with agent-assisted remote execution.
• Lab Teams seeking a shared, self-hosted research hub to maintain data control.
• Self-hosters who want full ownership of their infrastructure and model providers.

Highlights

• Grounded RAG Chat: Cited answers based on indexed local files.
• Paper Study Suite: Search, summarize, and run structured multi-role discussions on scientific papers.
• Deep Research: AI-driven multi-phase workflows with approval checkpoints and role-based execution.
• Scientific Skills Library: Over 200 built-in skills for domains like bioinformatics, genomics, and physics.
• Remote Execution: Orchestration of experiment loops via Shell, Slurm, and Kubernetes job schedulers.
• Multi-LLM Support: Compatible with OpenAI, Anthropic, Gemini, and various other providers.