In silico
Aether for drug discovery
30+ agents — docking, FEP, MD, QM, ADMET, pharmacophore, generative, biologics. The model reasons about molecules.
Aether for autonomous labs.
AI that closes the wet-lab loop.
Aether for autonomous labs is what an automated wet-lab looks like when the model is the substrate. Active-learning campaigns drive liquid handlers, acoustic dispensers and open-source robotics directly; chain-of-custody, ALCOA+ and 21 CFR Part 11 ship as part of the runtime. Ninety-two production modules cover the path from hypothesis to electronic batch record.
- Modules
- 92
- Compliance
- ALCOA+
- Instruments
- SiLA-2 + protocol export
- Audit
- 21 CFR 11
Capabilities
Six things this actually does.
Each capability below is a real module in the runtime, not a slide. Names map one-to-one to the production code paths.
01
Active-learning campaigns
Bayesian acquisition over chemical and biological space. Cost-aware: the agent picks the next experiment with the best information-per-dollar, not the highest model score.
02
Assay scheduling
Plate logistics, reagent-lot lineage, instrument allocation, cycle-time estimation. Gantt-grade scheduling with re-planning when an instrument falls over.
03
Instrument drivers
SiLA-2 gRPC for the major industrial liquid-handling platforms. Native protocol export for open-source robotics. Plate readers via standard adapters. Bench-top fixtures and custom rigs over a thin Python SDK.
04
Chain-of-custody
Every reagent, plate, sample and dilution — immutable history. eBR-grade, ALCOA+ compliant, 21 CFR Part 11 ready. Your auditor reads the trace directly.
05
Biosecurity
Dual-use refusals at the model layer, controlled-pathogen and select-agent lookups before a robot is allowed to start, export-control gating for sensitive workflows.
06
Failure memory
Every botched run is captured, classified and surfaced. The lab gets smarter; new hires inherit the lessons; the model uses past failures as priors.
The loop
One pass around the discovery loop.
Most platforms automate one step of discovery and leave the others to email. Aether automates the loop end-to-end — and then closes it.
- 01
Hypothesise
Propose targets, frame the design-of-experiments, score expected information gain against cost and time.
- 02
Plan
Bind reagents, schedule plates, allocate instruments, estimate cost and cycle-time. Re-planning is automatic when an instrument falls over.
- 03
Execute
Drive liquid handlers, plate readers and imagers — with full chain-of-custody for every dilution.
- 04
Read out
Image analysis, hit calling, ADMET endpoints, structure prediction, statistics. Curve fitting and IC50/EC50 are built in.
- 05
Decide
Update the model, pick the next experiment, write the eBR. Loop closed.
Instruments
Drives the bench your team already has.
First-class drivers for the instruments that show up in a real biotech. Bench-top fixtures and custom rigs over a thin Python SDK — no rip-and-replace.
Industrial 96/384-channel liquid handlers
Programmable pipetting workstations
High-throughput automated platforms
Open-source pipetting robotics
Acoustic-droplet dispensers (nL-scale)
Workstation automation + scheduler
Multi-mode plate readers
High-content imagers
Confocal screening imagers
Nanolitre droplet dispensers
Automated cell counters
Bench-top fixtures (custom)
What it replaces
The bespoke wet-lab automation stack today.
Most automated labs are a glued-together pile of vendor tools, custom Python and a LIMS that's older than the postdocs using it. Aether replaces the spine.
Vendor lab schedulers
Scientific-data lake platforms
Commercial scientific orchestrators
Commercial screening managers
Commercial ELN + data lakes
Off-the-shelf cell schedulers
Commercial cell-scheduling platforms
Bespoke LIMS automation
Compliance
Built for the regulated lab.
Compliance is not a bolt-on. The chain-of-custody, audit-trail and signature primitives live in the runtime — your auditor reads the same data the model uses to plan the next experiment.
ALCOA+
Attributable, legible, contemporaneous, original, accurate — and the four plus criteria. Auditor-grade.
21 CFR Part 11
Electronic records and signatures. Audit trails, role-based access, time-stamped immutability.
GxP-ready
GMP, GLP, GCP workflows with separation-of-duties controls and validated change management.
Biosecurity
Select-agent and controlled-pathogen gating before a tool call. Refusal at the model layer.
Vs the lab-automation stack
The whole loop, one model.
Cloud labs run the instruments and ELNs store the data, but none close the active-learning loop. Aether proposes, runs the plate and learns from the readout — under audit and biosecurity.
✓Active-learning campaigns
✓Liquid-handler / instrument drivers
✓Assay scheduling
✓ADMET / endpoint modelling
✓Chain-of-custody / eBR (ALCOA+)
✓Biosecurity guardrails
✓Protocol export (open robotics)
✓Closes the design→make→test loop
Why the cycle shortens
Months of campaign work, compressed to nights.
The slowest part of a wet-lab campaign isn't the assay — it's the planning, the queue, the analysis and the wait for the next round. Aether collapses each of those into the autonomous closed loop.
Stage
Today
With Aether
Why
Design a campaign
3–6 weeks
hours
The planner reads the brief, proposes the design-of-experiments, calls out instrument capacity and posts the plate map for your sign-off. No more committee meetings to spec a plate.
Run a round of plates
1–2 weeks (queue + ops)
overnight
Plate logistics, reagent-lot lineage and calibration-aware scheduling decide the order. Instruments run unattended while you sleep.
Analyse the results
days
minutes
The Apex world model reads the imaging directly — cell tracking, hit calling, segmentation — and feeds posteriors into active-learning without a CSV in between.
Plan the next round
1–2 weeks
same hour
Active-learning agents propose the next plate map as the previous round finishes. Decisions don't wait for Monday's modeling-team standup.
Continuous campaigns
24×7
Closed-loop active learning runs without human queue management. Lights-out wet lab, with a chain-of-custody log replaying every decision.
Plates per scientist
10×
Median throughput per scientist across customer pilots. Reclaimed time goes to the parts of the work only humans do well — biology, judgement, follow-up.
To a validated lead
−6 months
Programme-level compression on a typical small-molecule indication. The compounding gain is iteration count per quarter.
AI scientists, not chatbots
An autonomous bench scientist.
We're not selling lab orchestration. We're shipping a foundation model that reads the brief, schedules the plates, drives the instruments, reads the data and proposes the next round — under biosecurity guardrails it can't disable.
- 01
Reads the campaign brief
Ingests the objective, the assay menu, the SOPs, the biosafety boundary. Decomposes the brief into a plate-by-plate study tree.
- 02
Schedules the instruments
Plate logistics, reagent-lot lineage and calibration-aware scheduling — across liquid handlers, acoustic dispensers, plate readers and incubators.
- 03
Drives the bench
First-class SiLA-2 drivers for the major liquid handlers and dispensers. Protocol export for open-source robotics. Same runtime, your hardware.
- 04
Sees the result
The Apex world model reads the imaging directly. Cell tracking, hit calling, segmentation — calibrated against published datasets and your historical plates.
- 05
Updates the design
Active-learning agents propose the next plate map. Bayesian-optimal designs that move the posterior, not the most expensive ones.
- 06
Writes the eBR
21 CFR Part 11 electronic batch records auto-written as the campaign runs. Signed by the agent ID, the instrument ID and the model version.
Where this leads
A foundation model that runs the lab, not just plans for it — the substrate for general biological intelligence.
Aether sees plates, cells, signals and outcomes — the substrate of biology under a microscope. We are moving toward general intelligence through the disciplines that have to face physical reality. The model that runs your campaign tonight is the model that designs tomorrow's molecule.
In silico × on the bench
Aether proposes. The bench confirms.
Aether for drug discovery proposes the molecules and the experiments. Aether for autonomous labs binds reagents, schedules plates, drives the instruments, and feeds the measurements back. That is the loop. There is no email step.
On the bench
Aether for autonomous labs
Plate scheduling, instrument drivers, eBR, ALCOA+. Measurements flowing back into the discovery loop.
Related research
The work behind this product.
We publish what we learn. The posts below are the substantive notes behind this product — methods, evaluations, case studies.
Bring the bench into the model.
Tell us about your campaign. We'll show you what Aether does with one week of plates — and what your second week looks like.