Compute for AI agents

Compute that answers.

Your agent asks the question. OneShot computes the answer and shows the checks.

01 Solve

What is the best choice?

Routes, carrier awards, truck loads and shift rosters, solved to a proven bound.

Solve reference
  • RoutingVehicles, stops, capacities and time windows
  • AllocationLanes to carriers, lots to suppliers
  • PackingItems into bins by size and weight
  • SchedulingJobs on machines and people, in order
agent.solve({
  problem: 'allocation',
  allocation: {
    objective: 'minimise_cost',
    items: [{id: 'lane-1', quantity: 10}],
    options: [{id: 'carrier-a', capacity: 10}],
    pairs: [{item: 'lane-1', option: 'carrier-a', cost: 500}]
  }
})

02 Predict

What will happen?

Run a scenario thousands of times, or forecast a series. Every number comes with its range.

Predict reference
  • SimulationMonte Carlo runs on uncertain inputs, percentiles out
  • ForecastA time series in, prediction intervals out
agent.predict({
  model: 'simulation',
  simulation: {
    variables: [{ id: 'burn',
      distribution: { kind: 'normal', mean: 80, sd: 15 } }],
    outputs: [{ id: 'runway', expression: '900 / burn' }]
  }
})

03 Life sciences

What does the biology say?

Predict a protein's shape, dock a molecule, align sequences, and look up what is known.

Gene Summary reference

Models

  • Protein structureA 3D model from a sequence, scored per residue
  • DockingHow a molecule binds, poses ranked
  • Sequence alignmentAgainst each other or a reference set
agent.tool('lookup/gene-summary', {
  symbol: 'TP53'
})

Every sequence and molecule is screened before it runs.

Bring your sequences

Every answer carries its checks.

01Solve

A separate checker tests every constraint against the solution.

status
optimal
gap
to the proven bound
checks
every constraint, pass or fail

02Predict

Every number comes with its range and the checks that ran.

percentiles
p5 to p95
intervals
80% and 95%
checks
consistency of the inputs

03Life sciences

Every model answer says how sure it is and which model gave it.

confidence
per residue, per pose
model
name and version
screen
passed before it ran

Every paid call also returns a signed receipt.How receipts work

Your agent guesses.
Let it compute.