Skip to content

Fab Alpha preview

From engineering objective to evidence, faster.

Fab is where hardware teams run experiments, read telemetry and verify requirements. Dyno, its engineering agent, turns an objective into an investigation and links every finding to evidence you can open.

Application status See how it works

Application not ready yet.

Build. Simulate. Test. Learn.

DynoInvestigation trace
Illustrative workflowUnderstand rear instability.From an anomaly to a reasoned decision.
  1. OBSO-1301Rear ride-height oscillationFlagged
  2. HYPH-AFloor edge stalls at low ride heightSupported
  3. EXPE-253Floor edge +2 mm · heave ×1.15Complete
  4. RUNR-253CFD sweep · vehicle dynamicsSucceeded
  5. EVDEV-0418Observation linked to requirementStrong
  6. REQREQ-027Stall margin ≥ 5 mmPass
  7. DECDR-007Validate candidate on trackHuman draft
Concept preview · example identifiers and outcomes, not customer results.
Different physics. One engineering loop.MotorsportRoboticsFusion & fissionAdvanced hardware

The engineering loop

One system from question to decision.

Bring objectives, experiments, telemetry and reasoning into one connected record. The next experiment starts from what you already know.

  1. 01INV

    State an objective

    Define the question and its constraints. Keep the investigation connected to the system you are building.

  2. 02HYP

    Branch hypotheses

    Explore competing explanations in separate branches. Preserve what you learn when a hypothesis fails.

  3. 03EXP

    Run experiments

    Execute simulations, training jobs and physical tests through versioned capabilities on your compute.

  4. 04EVD

    Read the evidence

    Turn outputs and telemetry into observations, evidence and requirement evaluations with a traceable basis.

  5. 05DEC

    Decide and continue

    Compare candidates. A human records the decision, its alternatives and open risks; the next experiment starts here.

Dyno · investigate, test, advance

An engineering agent that shows its work.

Dyno plans branches, runs experiments through your capabilities, and reports findings you can trace. It executes autonomously through granted access, with activity and usage recorded alongside the investigation.

  • Evidence with a basis

    Strong, preliminary, inconclusive or an evidence gap. Each claim explains the runs, measurements or missing data behind it.

  • A clear next question

    Anomalies, thin margins and conflicting results become suggestions. Missing data stays visible instead of becoming a guess.

  • Human decisions, preserved

    You author decision records explicitly. Changed evidence marks dependencies for review; finalized decisions keep their original reasons.

DynoSynthesis · concept preview

The floor-edge candidate merits a track test. Here is what the example evidence supports.

Floor-edge stall explains the oscillation.

Strong

Basis: simulation sweep across ride heights.

Damping reduces amplitude in this scenario.

Preliminary

Basis: a small set of vehicle-dynamics runs.

Track behavior still needs evaluation.

Evidence gap

Missing: sufficiently sampled damper telemetry.

Next step: collect the missing measurement, then revisit the finding.
Candidate comparisonIllustrative values
Example simulation requirements, with units and margins
RequirementBaselineCandidate
Stall margin ≥ 5 mm4.4 mm · Fail5.6 mm · Pass
Rear load ≥ 1,450 N1,457 N · Pass1,466 N · Pass
Drag increase ≤ 1.5%Reference+1.2% · Pass
Source: illustrative simulation · a passing result does not replace measured validation.

Evidence over assertion

See the margin. Keep the reason.

A checkmark is only the beginning. Compare candidates against requirements and follow each result back to its run, configuration and source.

  • Predicted, simulated, measured

    Keep sources distinct. When they disagree, carry the conflict forward.

  • Alternatives and open risks

    A decision preserves what was considered, what remains uncertain and why you chose to proceed.

Engineering archetypes

One platform. Very different physics.

Example workflows for teams building systems in the physical world.

Motorsport / F1

Find an aero setup inside your rear-load limits.

Geometry → CFD → lap simulation → telemetry. Compare against a baseline, with requirement margins and configuration provenance.

Aero geometry · CFD · vehicle dynamics

Robotics

Improve inspection without sacrificing stability.

Explore sensing, mechanical and policy branches. Evaluate recall, coverage and fall rate across scenarios and seeds.

Robot configuration · simulation · evaluation

Fusion & fission

Explore physical constraints, one reproducible solve at a time.

Compare plasma equilibrium candidates against coil limits, or reactor parameter sweeps against thermal and safety constraints. Keep solver inputs and outputs attached.

Parameter sweep · solver artifacts · constraints

Your tools, your compute

Reproducible by design. Yours to run.

Fab is designed around versioned capabilities and a Kubernetes execution substrate. Bring solvers, simulators and training jobs into the same experiment loop.

Cloud, on-premises and air-gapped environments need explicit infrastructure, model and dependency planning. Deployment choices are part of your engineering setup.

Typed capabilitiesImmutable artifactsTelemetryKubernetesMCP

  • Run provenance

    Keep input hashes, image digests, configuration revisions and seeds with each run.

  • Access and usage provenance

    Track agent activity and usage alongside the capabilities and access granted to each execution.

  • Infrastructure boundaries

    Plan compute, data placement and disconnected dependencies for your environment.

  • Traceable outputs

    Link artifacts and telemetry to observations, evidence and requirements.

Fab Alpha preview

Application not ready yet.

Fab is still in development. The application is not available to use yet. This page previews the product we are building.