AutoResearch / 01Autonomous drug discovery

Run the question.Not the workflow.

Give AutoResearch a scientific goal. It designs the campaign, runs the compute, learns from the evidence, and returns with the record intact.

Bring us a target See the system
An interactive field representing hypotheses evolving through a research campaign.
Living search fieldMove / attractClick / seed
∞

Hypotheses stay in motion

StructureDockingADMETFEPDynamicsSelectivityRetrosynthesis+ five more

01 / The actual bottleneck

Discovery does not stall inside one model.It stalls between decisions.

AutoResearch turns the handoffs—strategy, design, compute, analysis, and reporting—into one continuous research object.

01 brief05 loop stages12 modalities01 trace

02 / The system

A brief enters. A campaign comes back.

Every stage produces something inspectable. Select a stage to see what the agent does, what your team controls, and what stays on the record.

Stage 01 / Research brief

Turn intent into a scientific contract.

The target, constraints, budget, stopping conditions, and definition of success become the campaign boundary.

Agent
Parses the goal and exposes ambiguity.
Scientist
Owns the question and signs the boundary.

03 / Research fabric

One operating system. A program’s worth of science.

Experiments stop behaving like separate tools. They share campaign memory, approval gates, compute, and provenance.

01

Structure prediction

Protein and complex structures become a live input to the campaign.

02

De novo docking

Generate candidates, score binding, evolve the next population.

03

ADMET

Pull developability into the loop before a lead becomes expensive.

04

FEP

Turn relative free energy into a decision signal.

05

Molecular dynamics

Test how promising systems behave beyond a single pose.

06—12

The rest of the program

Selectivity, retrosynthesis, crystal refinement, macrocycles, target ID, biomarkers, and patient stratification.

04 / Evidence ledger

Speed is useful.Knowing why is non-negotiable.

The campaign carries its own chain of custody: specialist debate, plan revisions, approvals, compute events, artifacts, and signed reports. The trace below is an illustrative example of that record.

ElapsedEventRecorded evidenceStatus
00:00BriefGoal, constraints, and success criteria receivedtraced
00:18CouncilThree specialist positions reconciledtraced
00:31PlanExecutable campaign package proposedtraced
00:44ApproveScientist gate recorded before computetraced
04:12RunGPU events and artifacts linked to decisionstraced
06:05ReportSigned result with full provenance emittedsigned
Human command / always

The agent can propose. Only your team can commit.

Campaign memory / durable

Every result changes what the system tries next.

05 / MoleculeLabs

Where foundation models meet real compute.

MoleculeLabs is building the research layer that turns powerful models into accountable scientific campaigns.

ML

Est. 2026

Your hardest question / our favorite starting point

Bring the target.We’ll show you the loop.

A technical conversation about your pipeline, your compute, and the experiment that should become autonomous first.

Request a technical briefing