Crucible · Autonomous labs
Runs the chemistry at scale.
A sample is not evidence until the chemistry has run. Crucible runs it at Coactive's labs in Ouray, Colorado and Reno, Nevada and in autonomous cells at national laboratories, and hands every result back to the model. The same active-learning rule that points the satellite and picks the sample designs every batch.
Where the loop ends
The loop does not end at a sample. It ends when the chemistry has run.
Composition, mineralogy, extractability, and the cost of each route. Crucible produces the numbers a buyer, a program, or a permit needs, and active learning decides which experiment to run next, the same way it decides where the satellite looks and which rock the field team picks up.
Ouray, Colorado
Coactive's characterization lab.
A crushing, pulverizing, and analysis line that takes samples from mine sites to a Passport, and screens extraction pathways with unit economics alongside the chemistry.
XRFQueued
ICP-OESQueued
PulverizerQueued
FurnaceQueued
Drying ovenQueued
CentrifugeQueuedPhotos: Ouray lab, 2026. Instrument readouts are illustrative.
Two loops on shared hardware
Characterize cheaply. Optimize the pathway.
Every sample gets the cheapest protocols first.
Intake, moisture, particle size, XRF, and a SWIR powder scan run first on every sample. Results update a belief about the sample, seeded by the orbital prior from Lodestar. A value-of-information ranker sends the sample to fusion and ICP-OES only when the expected information justifies the cost. The sample exits when its Passport fields reach target confidence.
Run candidate conditions in parallel. Rank the pathways.
Agents compile candidate leach and separation conditions from literature and internal methods. A liquid handler runs them on prepared sub-samples in parallel. Leachates go straight onto the ICP-OES, so there is no digestion bottleneck. The model scores recovery, selectivity, and unit economics per condition and designs the refinement round.
Output: ranked extraction pathways with cost per unit, in days and weeks instead of months.
Partner cells
Autonomous cells at national laboratories.
At PNNL, Meridian plans experiments for CICERO, a robotic chemistry workflow built on an Opentrons Flex liquid handler, with PNNL scientists running the chemistry and characterization. A parallel cell at NLR runs spectral and geochemical campaigns.
- PNNL
- CICERO. Closed-loop separations chemistry. Coactive supplies the decision layer that selects each batch by its value to the downstream route.
- NLR
- Spectral and geochemical campaigns. Sampled faces, hyperspectral signatures, and lab spectra tied back to orbital bands.
- INL
- Active sampling. Field sampling policy integration.
Contact
Bring a feedstock and a question.
What is in it, what is it worth, and what is the cheapest route to concentrate it. Crucible answers in rounds, not seasons.