Last updated: 11/20/2026Last verified: 2026-11-20
Ai2 scientific theory-generation tool that turns evidence from research papers into structured hypotheses with linked sources and scope conditions. Available through Asta and an Apache-2.0 Python library; web use has limits and local model calls cost extra.
-
Generate scientific hypotheses from evidence extracted across research papers
-
Inspect structured laws, scope conditions, and references supporting each claim
-
Use the Generate Theories tab in Asta with a free account and capped sponsored usage
-
Run the Apache-2.0 Python library locally with your own papers and paid model API keys
Details
Platform
WebLinux
Features
- Generate scientific hypotheses from evidence extracted across research papers
- Inspect structured laws, scope conditions, and references supporting each claim
- Use the Generate Theories tab in Asta with a free account and capped sponsored usage
- Run the Apache-2.0 Python library locally with your own papers and paid model API keys
Languages
en
Best for
Scientists developing evidence-linked explanations for unexpected observationsResearchers synthesizing patterns across open-access scientific literatureDevelopers adapting an open-source literature-to-theory workflow to their own papers
Known limitations
- Generated theories can be misleading, miss relevant studies, or misjudge originality; researchers must verify evidence and test hypotheses
- Asta requires a free account for theory generation and covers token costs only up to a limit; web retrieval is restricted to open-access papers
- Local execution needs Python 3.12+ and your own paid model API keys; large literature runs can incur significant token costs
- The local installation is tested on Ubuntu Linux; macOS and Windows may require modifications
Alternatives
NotebookLM — Source-grounded document research and summaries rather than structured scientific theory generation
Deep Review — Reviewing technical arguments and literature in an existing research manuscript
Cloudy — Running machine-learning experiments instead of synthesizing theories from papers







