Agent-generated screening
AI generates candidates, platform governs each independently. Campaign-level audit trails.
Examples
Applied patterns for authoring .matr programs across domains. Each example demonstrates real constructs, governance annotations, and constraint patterns that map into Material Cloud execution profiles. These are not toy programs. They represent the kinds of Material Computing workloads the platform is designed to handle.
Enterprise AI: molecular screening agent
An AI agent generates candidate matter programs, submits them to the Material Twin, evaluates signals, and iterates. This pattern is common in drug discovery, materials optimization, and any domain where an agent explores a large parameter space under governance constraints.
// screening-candidate-0147.matr
// Generated by: BERNIE screening agent v3.2
// Campaign: small-molecule-binders-2026Q1
import { TargetProtein } from "@org/targets/HER2-v4"
import { StandardAssay } from "@org/protocols/binding-assay"
material Candidate0147 {
substrate: "synth-peptide-lib"
sequence: "CYCRGDFC"
molecular_weight: 912.04 Da
purity_required: >= 0.90
}
protocol BindingScreen {
step synthesize {
input: Candidate0147
method: "solid-phase"
duration: 4 hr
temperature: 25°C
}
step assay {
input: step.synthesize.output
target: TargetProtein
protocol: StandardAssay
concentration_range: [1 nM, 10 uM]
replicates: 3
}
}
constraints {
binding_affinity <= 100 nM // Kd threshold
selectivity_ratio >= 10 // vs off-targets
synthesis_yield >= 0.70
assay_z_score >= 0.5
}
governance {
tier: T1
consequence: "physical-negligible"
approvals: ["auto"]
campaign: "small-molecule-binders-2026Q1"
agent: "bernie-screening-v3.2"
}
simulation {
runs: 500
confidence: 0.90
substrate_model: "synth-peptide-lib-v2"
baseline: "campaign-best-so-far"
}
output {
binding_affinity: Kd in nM
selectivity_ratio: ratio
synthesis_yield: fraction
assay_z_score: score
}Agent-generated programs with campaign-level governance. The agent iterates; the platform governs each iteration independently.
T1 with auto-approval. Low consequence per run, but campaign-level audit trails link every candidate to the generating agent.
The agent reads signals from each run to decide the next candidate. Baseline comparison tracks progress across the campaign.
Scale biology: parallel batch synthesis
A production workflow that synthesizes nanoparticle batches at scale. Multiple protocols compose into a pipeline, with constraints that tighten at each stage. This pattern is used when physical parallelism across substrate pools is required to meet throughput targets.
// nanoparticle-batch-production.matr
material GoldSeed {
substrate: "colloidal-au-production"
diameter: 5 nm ± 0.5 nm
concentration: 2.0 mM
lot: "AU-2026-0215"
}
material CitrateBuffer {
substrate: "reagent-stock"
compound: "trisodium-citrate"
concentration: 34 mM
volume: 500 mL
}
protocol GrowthPhase {
step nucleation {
input: [GoldSeed, CitrateBuffer]
temperature: 100°C
duration: 30 min
stir_rate: 600 rpm
atmosphere: "N2"
}
step growth {
input: step.nucleation.output
temperature: 100°C → 70°C ramp over 45 min
duration: 2 hr
stir_rate: 300 rpm
}
step quench {
input: step.growth.output
temperature: 4°C
duration: 15 min
method: "ice-bath"
}
}
protocol Characterization {
step dls {
input: GrowthPhase.output
method: "dynamic-light-scattering"
replicates: 5
}
step tem {
input: GrowthPhase.output
method: "transmission-electron-microscopy"
sample_count: 200
}
step zeta {
input: GrowthPhase.output
method: "zeta-potential"
replicates: 3
}
}
constraints {
particle_diameter: 50 nm ± 5 nm
polydispersity_index < 0.15
zeta_potential: -40 mV ± 10 mV
batch_yield >= 0.85
endotoxin < 0.25 EU/mL
batch_to_batch_cv < 0.08
}
governance {
tier: T2
consequence: "physical-reversible"
approvals: ["production-lead", "qa-review"]
escalation: "halt-and-notify"
compliance: ["GMP", "ISO-13485"]
lot_tracking: true
}
simulation {
runs: 2000
confidence: 0.95
substrate_model: "colloidal-au-production-v5"
baseline: "batch-AU-2026-0201"
timeout: 600s
}Multi-protocol composition with characterization as a separate stage. The platform schedules growth and characterization as linked runs.
T2 with production and QA approval. GMP compliance and lot tracking are enforced at the language level.
Batch parallelism is handled by Material Cloud facility orchestration, not the program. The same Molebyte executes across available facility capacity.
Precision sensing: environmental biosensor
A high-precision program for fabricating and calibrating a biosensor array. Constraints are tight, governance is elevated, and simulation parameters require high confidence. This pattern applies to diagnostics, environmental monitoring, and any application where measurement accuracy is critical.
// biosensor-array-v2.matr
material AptamerProbe {
substrate: "oligo-synth"
sequence: "GGTTGGTGTGGTTGG" // thrombin aptamer
modification_5: "thiol-C6"
modification_3: "FAM"
purity_required: >= 0.98
}
material GoldElectrode {
substrate: "electrode-fab"
surface: "Au(111)"
area: 0.196 cm²
roughness_factor: 1.2 ± 0.1
}
protocol Fabrication {
step clean {
input: GoldElectrode
method: "piranha-etch"
duration: 5 min
temperature: 25°C
safety: "fumehood-required"
}
step immobilize {
input: [step.clean.output, AptamerProbe]
method: "self-assembled-monolayer"
concentration: 1 uM
incubation: 16 hr
temperature: 4°C
atmosphere: "N2"
}
step backfill {
input: step.immobilize.output
reagent: "mercaptohexanol"
concentration: 1 mM
duration: 1 hr
temperature: 25°C
}
step calibrate {
input: step.backfill.output
analyte: "thrombin"
concentration_series: [0.1 nM, 1 nM, 10 nM, 100 nM, 1 uM]
method: "square-wave-voltammetry"
replicates: 5
}
}
constraints {
probe_density: 4e12 molecules/cm² ± 1e12
signal_to_noise >= 10
limit_of_detection <= 0.5 nM
linear_range: [1 nM, 100 nM]
coefficient_of_variation < 0.05
electrode_to_electrode_cv < 0.10
}
governance {
tier: T3
consequence: "physical-significant"
approvals: ["sensor-lead", "metrology", "safety-review"]
escalation: "halt-review-revalidate"
compliance: ["ISO-15197", "IVD-directive"]
}
simulation {
runs: 5000
confidence: 0.99
substrate_model: "electrode-fab-v4"
baseline: "biosensor-array-v1-production"
timeout: 900s
}Precision fabrication with tight tolerances and multi-step surface chemistry. Each step builds on the previous, with no room for out-of-spec intermediates.
T3 with three-person approval chain. IVD compliance labels are embedded in the artifact and enforced by Material Cloud policy.
5,000 simulation runs at 99% confidence. The high bar reflects the cost and irreversibility of electrode fabrication failures.
Hybrid AI: closed-loop optimization
A pattern where AI-generated matter programs and Material Cloud execution form a closed loop. The agent proposes formulations, Material Cloud executes them, signals feed back to the agent, and the agent generates improved programs. Governance escalates automatically as the agent approaches high-value parameter regions.
// polymer-optimization-iter-023.matr
// Generated by: BERNIE formulation agent v2.1
// Loop: closed-loop-polymer-opt-2026
// Iteration: 23 of max 100
import { BasePolymer } from "@org/materials/PLGA-5050"
import { Surfactant } from "@org/materials/PVA-31k"
material FormulationIter023 {
substrate: "nanoparticle-encapsulation"
polymer: BasePolymer
surfactant: Surfactant
// Agent-optimized parameters for this iteration
polymer_concentration: 12.5 mg/mL // was 10.0 in iter-022
surfactant_concentration: 2.1% w/v // was 2.5% in iter-022
organic_phase: "acetone"
aqueous_phase: "water"
drug_loading_target: 8% w/w
}
protocol Nanoprecipitation {
step prepare_organic {
input: FormulationIter023
method: "dissolve"
solvent: "acetone"
duration: 30 min
temperature: 25°C
}
step precipitate {
input: [step.prepare_organic.output, Surfactant]
method: "dropwise-addition"
flow_rate: 0.5 mL/min
stir_rate: 800 rpm
temperature: 25°C
}
step evaporate {
input: step.precipitate.output
method: "rotary-evaporation"
temperature: 40°C
pressure: 150 mbar
duration: 45 min
}
step characterize {
input: step.evaporate.output
measurements: ["DLS", "zeta", "encapsulation-efficiency"]
replicates: 3
}
}
constraints {
particle_size: 150 nm ± 30 nm
polydispersity_index < 0.20
encapsulation_efficiency >= 0.60
drug_loading >= 0.06 // 6% w/w minimum
zeta_potential < -20 mV
}
governance {
tier: T1
consequence: "physical-negligible"
approvals: ["auto"]
escalation: "notify-if-constraint-fail"
loop: "closed-loop-polymer-opt-2026"
iteration: 23
agent: "bernie-formulation-v2.1"
// Auto-escalates to T2 if agent approaches
// high-value parameter region (defined in policy)
auto_escalation: {
condition: "encapsulation_efficiency >= 0.85"
target_tier: T2
reason: "approaching production-viable formulation"
}
}
simulation {
runs: 200
confidence: 0.90
substrate_model: "nanoparticle-encapsulation-v3"
baseline: "iter-022-signals"
}Closed-loop optimization where each iteration is a governed program. The agent controls parameters; the platform controls governance.
Governance tier escalates automatically when the agent finds high-performing formulations. This prevents unsupervised crossing into production-viable territory.
Every iteration is linked: agent version, loop ID, iteration number, and baseline reference create a complete optimization history in Vault.
Pattern summary
Multi-stage production
Composed protocols with tightening constraints at each stage. Facility-level parallelism.
Precision fabrication
Tight tolerances, elevated governance, high-confidence simulation. For diagnostics and metrology.
Closed-loop optimization
Agent-driven iteration with auto-escalating governance as results improve.