Neutropic Beta
Your research agent for reliable science
The Neutropic app runs analyses, searches databases, and traces every step from data processing to publication, so you can spend time on research.
Built for scientific research
Rich research artifacts, fully reproducible
Research is inherently visual. Neutropic generates figures, tables, and reports alongside the code that produced them — annotate any artifact and revise it in plain language. Every output carries its data source, random seed, and environment, so anyone can reproduce it.
- Figures, tables, and reports saved with their generating code and data source
- Built-in renderers for SEM path diagrams, gaze heatmaps, and signal time series
- A review agent checks statistical assumptions, multiple comparisons, and power
- Comment on any artifact to iterate on figures in natural language
- Turn your analysis into a manuscript draft
Manages your compute and scales on demand
Install the app and it runs on localhost, so sensitive data — survey responses, recordings, biosignals — never leaves your machine. Large datasets load once into session memory, and heavy work scales out only when you need it.
- Persistent Python and R kernels across a session
- Load large surveys, logs, and signals once per session
- Scale heavy signal processing and simulations to multiple cores or opt-in remote
- Consistent environments across devices
Domain-ready on day one
Skills, connectors, and literature databases come pre-configured for psychology, emotion engineering, and HCI. Connect the models, datasets, and pipelines your lab already trusts, then save them as reusable skills.
- Skill bundles for psychometrics, emotion engineering, multimodal signals, and HCI
- Connectors for Qualtrics, REDCap, OSF, PsychoPy, and iMotions
- Literature search across PsycINFO, PubMed, Scopus, and Semantic Scholar
- Save your lab's pipelines as custom, shareable skills
How researchers use Neutropic
Factor analysis, SEM, reliability, and power — through conversation
* The showcase is filled with real domain-skill outputs (placeholders for now). We never publish fabricated results.