Home/AI for Science/OpenScience
Product

OpenScience

Open-source, model-agnostic AI workbench that runs the full scientific research loop — literature → hypothesis → code → experiment → analysis → write-up — on your own keys and infrastructure.

1. Core Product / Service

OpenScience (openscience.sh) is an Apache-2.0 AI workbench from Synthetic Sciences, released July 2026 [1][2]. It runs the complete scientific research loop — literature, hypothesis, code, experiments, and the write-up — in one browser-based workspace, on any model the user points it at, with their own API keys and infrastructure [2].

Local research (2026-09) documented its architecture:

  • 39 non-literature science databases integrated (UniProt, PDB, ChEMBL, KEGG, AlphaFold, and others) via a connector framework.
  • Real code execution + experiment loop — dry-lab simulation on a 7-state machine; wet-lab via natural-language → Opentrons protocols.
  • Skills / tools / connectors — a modular layer where skills wrap capabilities and connectors bind data sources.
  • Model-agnostic — no lock-in to a single LLM provider.

2. Target Users & Pain Points

  • Researchers / labs who want a reproducible, self-hostable research agent without vendor lock-in.
  • Builders benchmarking against an open reference architecture.
  • Pain solved: commercial tools are closed and managed; OpenScience offers the open, BYO-infrastructure alternative.

3. Competitive Landscape

Tool Type Note
OpenScience Open-source, model-agnostic workbench Apache-2.0, self-host
claude-science Claude Science Managed workbench (Anthropic)
aminer AMiner Academic search + graph

OpenScience's edge is openness plus breadth of scientific databases (39), versus Claude Science's managed polish and model integration.

4. Unique Observations

  • The open pole of the agentic-science race. As claude-science commercializes "AI workbench for scientists," OpenScience is the Apache-2.0 counterweight — model-agnostic and self-hosted, which matters for labs with data-residency or cost constraints.
  • Workflow breadth, not model strength. Its differentiation is the loop — literature + code + experiment + write-up in one place — rather than any single model; users plug in whichever LLM they prefer.
  • Local gap analysis (2026-09) found OpenScience lacks AMiner's scholar profiling, citation-network, patent search, and Chinese-literature coverage — its 39 databases skew toward life-science/chemistry data rather than academic-graph features [local].

5. Financials / Funding

Open-source project published by Synthetic Sciences; no disclosed funding. (Synthetic Sciences is the GitHub org; company details unverified in local research.)

6. People & Relationships

  • Publisher: Synthetic Sciences (github.com/synthetic-sciences).
  • Ecosystem: claude-science (closed counterpart), aminer (academic data source it does not fully cover).
Last compiled: 2026-09-14