An AI-powered scientific research automation platform for materials science, covering the full research cycle from experimental design to data interpretation through planning, analysis, and simulation agent systems.
SciLink is an AI-powered scientific research automation platform for materials science, developed and maintained by Maxim Ziatdinov. Currently at version 0.0.26 (Beta), the platform covers the full research cycle through three complementary agent systems:
Planning Agents handle automated experimental design, workflow orchestration, data scalarization, and Bayesian parameter optimization, leveraging a dual knowledge base (docs + code) and RAG engine for literature-aware experiment planning. Analysis Agents support multimodal data analysis — 1D curve fitting (XRD, UV-Vis, PL, XPS, Raman, etc.), microscopy image analysis (SEM, TEM, AFM, STEM with two-tier pipeline: basic detection and deep analysis including strain mapping and sublattice separation), and hyperspectral data cube processing (EELS-SI, EDS, Raman imaging), with automatic scientific claim generation and novelty assessment via FutureHouse AI. Simulation Agents (coming soon) are planned to cover DFT workflow automation and LAMMPS-based molecular dynamics simulation.
The platform offers three autonomy levels (Co-Pilot / Supervised / Autonomous), bidirectional MCP protocol support (as Server exposing tools to clients like Claude Code, and as Client connecting to external MCP servers), and plugin-based extensibility for custom tools, skills, and agents via the entry-points mechanism. Under the hood, it uses LiteLLM for unified multi-LLM interfaces, the author's own atomai library for atomic-level image analysis, lmfit for curve fitting, BoTorch for Bayesian optimization, and FAISS for vector retrieval. It provides CLI, Python API, and Streamlit Web UI interaction modes, with Docker containerization support.