Generalizable Experimental Methods and Standards for Transparent, Open, Networked Execution

Building America’s AI infrastructure for programmable cloud laboratories

GEMSTONE is a node in the NSF PCL testbed network of programmable cloud laboratories.

Scientists should be able to describe an experiment in a manner that is natural to them — in prose — and to have their experiment run reliably and reproducibly in any automated laboratory. GEMSTONE is building the open standards and AI agents that make this approach possible — starting at Emerald Cloud Lab and setting the pattern for a national network of programmable cloud laboratories.

Read about the approach Get in touch

740+

programmable instruments

320+

unique instrument types

100,000

sq ft computer-controlled facility

partner organizations

A long aisle of the automated laboratory floor at Emerald Cloud Lab, lined with incubator shakers whose displays show live speed and temperature readings, receding toward scientists working in the distance. — Incubator shakers running mid-experiment on the laboratory floor in Austin — one aisle among rows spanning the 100,000 sq ft facility.

Four glass reagent bottles fitted with sensor caps and tubing, each monitored by its own dedicated machine-vision camera. — Even the bottles are instrumented: sensor caps and dedicated cameras track the state of every vessel.

A rack labeled 'Pipette Station' holding dozens of pipettes, each tagged with its own QR-code label. — Every asset down to the individual pipette is barcoded, so software can locate, schedule, and audit each one.

From prose to execution

A major bottleneck in automated experimentation is method transfer: writing a specification that is sufficiently detailed to guide a programmable cloud laboratory reliably. Researchers are not trained to write out the myriad, low-level, common-sense details that machines require for workflow execution. GEMSTONE places an abstract, node-independent layer between natural language descriptions of experiments and the platform-specific languages that drive instruments and that enable efficient, fully automated workflows.

Input

Prose protocol

A USP monograph, a journal methods section, or a researcher’s own description.

ORE

Disambiguation

An AI agent flags vague instructions — “mix thoroughly,” “adjust pH as needed” — and resolves them by asking the user or by running small experiments.

AMBER

Representation

The Abstract Model for Bridging Experiment Representations: human-readable for validation, machine-executable for automation.

TROVE

Translation

A second agent converts AMBER into a node’s own language, beginning with Emerald Cloud Lab’s Symbolic Lab Language.

Execution

The node runs it

Robotic handling, instrument runs, and provenance capture inside a 100,000 sq ft facility in Austin.

Closed loop

Results feed back

Outcomes return to TROVE, which adjusts parameters, updates constraints, and proposes improved variants.

FAIR data

Data archiving

All data generated through our approach are described using standards-adherent metadata and submitted to appropriate repositories, ensuring that datasets are findable, accessible, interoperable, and reusable.

Partners

One distributed laboratory spanning five organizations. Select a site to see what it contributes.

Stanford University

Data standards and the AMBER representation

Leads the design of AMBER and the tools that edit and maintain it, building on CEDAR, BioPortal, and Protégé.

Part of the GEMSTONE team, in lab coats and safety glasses, standing together on the automated laboratory floor at Emerald Cloud Lab in Austin, Texas. — Part of the GEMSTONE team on the laboratory floor at Emerald Cloud Lab in Austin, Texas, during the project kickoff, August 25–26, 2026.

Open by design

GEMSTONE is laying the shared, interoperable foundations for a national network of programmable cloud labs — open infrastructure that any researcher, institution, or lab can build on.

An open standard

AMBER is an open, implementation-agnostic standard — meant to do for laboratory automation what STEP did for manufacturing: a shared interchange layer before each node becomes too siloed to interact with the others.

Open-source execution

Emerald Cloud Lab’s Symbolic Lab Language is contributed to the Test Bed under an open-source license.

Open artifacts

Trained models, AMBER templates, evaluation harnesses, and training data are shared under open licenses through a public GitHub repository.

FAIR-compliant

AMBER specifications are versioned, provenance-tracked, and shared in open formats, so other nodes can inspect, re-execute, and extend workflows.

Latest

All news →

Aug 2026

GEMSTONE kicks off in Austin

The GEMSTONE team gathered in Austin, Texas, on August 25–26, 2026, for the project’s kickoff meeting, hosted by Emerald Cloud Lab — the partner whose automated facility will ru...