Hit target density, geometry and takt before the build starts.
Parttwin is the as-built part digital twin.
Products
Buy the stage that hurts. Add the rest when it pays.
Powdrix ships as five products mapped to the additive floor, not to abstract capabilities. Parttwin plans the build, Meltrix executes it at the melt pool, Poroscan verifies it at CT, Forgebot handles plates, powder and post-processing, and Yieldex runs the fleet. Start with one on the Machine tier and expand as each proves its number.
Illustrative scenario: PBF-2 · Ti-6Al-4V bracket family, 24-part build, 23 of 24 parts released, one held
The five
One stack, five products, one genealogy
Parttwin simulates the build and hands a validated recipe to Meltrix. Meltrix holds the melt pool and emits the build-health stream. Poroscan grades every part at CT and links defects back to the signal. Forgebot moves plates, powder and parts and runs post-processing. Yieldex predicts failure, schedules the fleet and owns genealogy and conformance.
See and steady every melt pool, layer by layer.
Meltrix is the printer-edge perception-and-control core.
Every pore, crack and micron of geometry, qualification-grade.
Poroscan is the inspection-and-qualification engine.
Autonomous build-plate, powder and part handling through post-processing.
Forgebot is the physical-AI handling-and-post-processing layer: Isaac and Jetson driven robots that load and unload build plates, handle and sieve powder, depowder parts and transfer them through heat treatment, HIP and support removal, coordinated with live machine state from Meltrix and sequenced by Yieldex.
The floor brain: yield, takt and qualification across the fleet.
Yieldex is the floor-level optimisation-and-conformance brain.
Parttwin · brand #2563EB · design stage
Parttwin
Hit target density, geometry and takt before the build starts.
The problem
Build success is decided before the first layer, by orientation, supports, nesting and the process-parameter recipe. Getting them right is craft-bound trial and error: engineers burn scrapped multi-hour builds and coupon campaigns to find a recipe that avoids distortion, lack of fusion and support failure while hitting density, geometry and takt. Fixed parameters and manual prep cap yield, and the scarce process-engineering workforce cannot explore the design space fast enough to scale the fleet.
What Parttwin does
Parttwin is the as-built part digital twin. It simulates the build, thermal fields, a melt-pool proxy, residual stress and distortion, for candidate orientation, support, nesting and parameter recipes, and auto-optimises them against density, geometry, takt and energy targets before the build is released. It generates rare defect and distortion scenarios to stress-test recipes and hands the winning, validated recipe to Meltrix for execution.
Who it is for
Aerospace, medical, energy, defence and tooling AM producers and service bureaus doing build preparation and parameter development, and AM engineering teams scaling qualified part families across a machine fleet.
Key features
- GPU thermal, melt-pool-proxy, residual-stress and distortion build simulation
- Auto-optimisation of orientation, supports, nesting and process parameters to targets
- Synthetic rare-defect, distortion and recoat scenario generation for recipe stress tests
- As-built versus as-designed density and geometry prediction with confidence
- Validated recipes emitted to Meltrix and predicted-defect maps to Poroscan
- Calibration against Poroscan CT results as builds complete
NVIDIA stack
Named in the product brief as the current design
OmniversecuOptCosmos and Omniverse ReplicatorDGX or OVXH100 or A100 trainingRTX workstationsNIM
Planned
Modulus surrogates · plannedAlloy-specific surrogate foundation models · planned
Why it needs a GPU
Thermal, distortion and residual-stress simulation and thousand-scenario parameter sweeps per part family are massively parallel workloads. CPU-only solves take days and cannot support pre-build optimisation at production cadence.
MVP scope and first metric
Simulate one part family's thermal and distortion for a few candidate recipes, rank against density, geometry and takt, and export the best recipe to Meltrix. Calibrate against Poroscan CT results as builds complete.
Tier. Sold inside the Floor tier ($70,000 per month) and Enterprise recipe and qualification programmes.
Open questions
- Simulation fidelity versus runtime trade-offs per alloy and process are not yet characterised
- Transferable alloy-specific surrogate models are planned, not built
Meltrix · brand #F97316 · design stage
Meltrix
See and steady every melt pool, layer by layer.
The problem
Metal powder-bed fusion prints a part from thousands of melt pools per layer across hours-to-days-long builds. A melt-pool instability, keyhole event, lack of fusion or recoat-linked thermal excursion buried at layer 2,000 becomes a hidden pore or crack: a scrapped multi-hour build, a failed qualification or a safety-critical field escape. Floors run fixed parameters, review in-situ data after the fact, and pace everything on a scarce process engineer who cannot watch terabytes of melt-pool video in real time.
What Meltrix does
Meltrix is the printer-edge perception-and-control core. It fuses coaxial and off-axis melt-pool video, layer and recoater images, photodiode and thermal streams at high frame rate, classifies and predicts porosity, lack of fusion, keyholing, spatter and recoat faults seconds ahead, and closes into bounded, operator-graduated laser and energy control to hold the pool stable. It emits a live build-health stream every other Powdrix product consumes, tagged to immutable build genealogy.
Who it is for
Aerospace, medical, energy, defence and tooling AM producers and service bureaus running laser or electron-beam powder-bed fusion and directed energy deposition, and multi-machine AM fleets.
Key features
- High-frame-rate melt-pool, layer and recoater defect classification and prediction
- Bounded closed-loop laser power, speed and focus stabilisation inside the approved envelope
- Multimodal sensor fusion, vision, photodiode and thermal, at the printer edge
- Layer-and-genealogy-linked defect logging with human-checkpointed actions
- A common build-health signal for Poroscan, Parttwin, Forgebot and Yieldex
- Retrofit onto existing coaxial melt-pool and layer cameras
NVIDIA stack
Named in the product brief as the current design
Jetson OrinHoloscanDeepStream and MetropolisTensorRTTritonNIMDGX or HGX training
Planned
Jetson Thor · plannedCosmos rare-event synthesis at scale · plannedAlloy-transfer of rare-event models · planned
Why it needs a GPU
Real-time inference on 30 to 240 fps melt-pool, layer and thermal streams per printer with sub-250 ms risk latency is not possible on CPU. Training on millions of high-resolution frames plus synthesised rare lack-of-fusion and keyhole variants needs DGX or HGX class capacity.
MVP scope and first metric
Retrofit one metal PBF printer with its existing coaxial melt-pool and layer cameras. Ship defect classification and lack-of-fusion risk alerts as advisory, then graduate to bounded laser and energy control feeding the rest of the loop.
Tier. Machine tier, $6,000 per AM machine per month, expanding to Floor and Enterprise.
Open questions
- Sensor suites, frame rates and OEM control-API access vary by machine
- Alloy transfer of rare-event models waits on customer data agreements
Poroscan · brand #0EA5E9 · design stage
Poroscan
Every pore, crack and micron of geometry, qualification-grade.
The problem
A printed part is only sellable once it passes inspection: CT for internal porosity, cracks and lack of fusion, vision for surface faults, metrology for geometry against spec. Today this is spot CT on a sample, hours of manual review and terabytes of un-interpreted volumes. The bottleneck gates qualification, hides field-escape risk, cannot keep pace with serial production and never links a defect back to the build that caused it.
What Poroscan does
Poroscan is the inspection-and-qualification engine. GPU 2D and 3D segmentation models triage CT volumes for pores, cracks and lack of fusion, fuse vision surface inspection and metrology deviation, grade each part against the alloy and part acceptance criteria with a human sign-off checkpoint, and assemble a citable qualification packet tied to immutable build genealogy. Every defect is linked back to the Meltrix in-situ signal and the Parttwin prediction.
Who it is for
Aerospace, medical-implant, energy and defence AM producers with qualification obligations, service bureaus offering certified parts, and CT and metrology OEMs embedding Powdrix analysis.
Key features
- GPU 2D and 3D CT segmentation of porosity, cracks and lack of fusion
- Vision surface-defect and metrology geometry-deviation fusion
- Criteria-aware accept and reject grading with human sign-off checkpoints
- Citable qualification and conformance packets bound to build genealogy
- Correlation of as-built defects to Meltrix precursors and Parttwin predictions
- Standards-aware grading rationale with mandatory citations
NVIDIA stack
Named in the product brief as the current design
L4, L40S or A100 inferenceTritonTensorRTCUDA and cuDNNNIMNeMo Retriever and GuardrailsDGX or HGX training
Planned
Active-learning scan prioritisation · plannedIn-situ-to-CT correlation learning · planned
Why it needs a GPU
3D CT segmentation of multi-gigabyte volumes at production cadence, plus training on large annotated volume corpora and synthetic rare-defect variants, is intractable on CPU: hours per part versus minutes. GPU acceleration is the difference between a gate and a bottleneck.
MVP scope and first metric
Connect one CT and metrology cell for a single qualified part family. Ship automated porosity and lack-of-fusion triage with accept and reject suggestions as advisory, with citable packets, then graduate to automated grading gates and in-situ correlation.
Tier. Machine tier, $6,000 per machine or inspection cell per month, expanding to Floor and Enterprise qualification programmes.
Open questions
- CT modalities, resolutions and acceptance criteria vary by part and standard
- Active-learning scan prioritisation and in-situ to CT correlation are planned
Forgebot · brand #10B981 · design stage
Forgebot
Autonomous build-plate, powder and part handling through post-processing.
The problem
Around every AM machine sits dangerous, manual, engineer-bound physical work: loading and removing heavy build plates, handling and sieving reactive metal powder, depowdering parts, and moving them through stress relief, heat treatment, HIP and support removal. It is hazardous, labour-scarce and a takt bottleneck, so machines sit idle between builds and post-processing runs blind and reactive.
What Forgebot does
Forgebot is the physical-AI handling-and-post-processing layer: Isaac and Jetson driven robots that load and unload build plates, handle and sieve powder, depowder parts and transfer them through heat treatment, HIP and support removal, coordinated with live machine state from Meltrix and sequenced by Yieldex. It controls oven and HIP recipes and support removal, keeping the physical loop moving safely and on takt with full genealogy tracking.
Who it is for
AM producers and service bureaus with multi-machine floors seeking lights-out plate, powder and part handling and automated post-processing, and automation integrators building AM cells.
Key features
- Robotic build-plate load and unload, powder handling and sieving, depowdering
- Vision-guided grasping and safe motion coordinated with printer and oven state
- Post-processing control: stress relief, heat treatment, HIP and support-removal recipes
- Genealogy-tracked part transfer from print to depowder to post-process to inspection
- Coordination with Meltrix build state and execution of Yieldex fleet schedules
- Reactive-powder-safe motion planned and checked in simulation first
NVIDIA stack
Named in the product brief as the current design
Jetson OrinIsaac ROS and Isaac ManipulatorIsaac Sim and Isaac LabMetropolis and DeepStreamcuOptNIMDGX or A100 training
Planned
Jetson Thor · plannedFleet Command edge management · plannedAdaptive post-process recipe control · planned
Why it needs a GPU
Real-time on-robot perception, grasp planning and safe motion, plus sim-to-real training with heavy domain randomisation, are GPU-bound. CPU-only cells cannot meet the latency or train robust policies for reactive-powder handling.
MVP scope and first metric
Automate one handling step, build-plate load and unload or depowdering, on a single cell with Isaac-trained grasping, coordinated with Meltrix build-complete signals. Expand to powder handling and post-processing control.
Tier. Sold inside the Floor tier ($70,000 per month) plus per-cell handling modules, and Enterprise multi-site fleet contracts.
Open questions
- Cell layouts, grippers and oven and HIP interfaces vary by floor
- Sim-to-real transfer for reactive-powder handling is planned and unproven
Yieldex · brand #8B5CF6 · design stage
Yieldex
The floor brain: yield, takt and qualification across the fleet.
The problem
An AM floor is a fleet of expensive machines, ovens, HIP units, inspection cells and robot cells, each producing signals no one can watch together. Managers run blind on real-time yield, build-failure risk, energy, takt and build genealogy, scheduling is manual, and every qualification demands traceable evidence assembled by hand. Build failures, idle machines, energy waste and non-conformances erode cost per part and stall the serial-production shift.
What Yieldex does
Yieldex is the floor-level optimisation-and-conformance brain. It ingests build health from Meltrix, inspection results from Poroscan, recipes from Parttwin and cell state from Forgebot, predicts build-failure and qualification risk, schedules machines, ovens, HIP, inspection and robots for yield, takt and energy with cuOpt, and maintains an immutable, standards-aware build-genealogy and conformance record with cited AM-process reasoning over the whole fleet.
Who it is for
AM production managers, quality and conformance leaders and operations teams at multi-machine producers, service bureaus and multi-site manufacturers running qualified serial production.
Key features
- Fleet-wide build-failure and qualification-risk prediction from multimodal floor data
- cuOpt scheduling of machines, ovens and HIP, inspection, robots and energy windows
- Immutable build genealogy and conformance record with standards-aware, cited reasoning
- Yield, takt, energy and right-first-time views and non-conformance workflows
- Orchestration of Meltrix, Poroscan, Parttwin and Forgebot as one floor system
- Re-optimisation after build delays or quality holds
NVIDIA stack
Named in the product brief as the current design
cuOptRAPIDSTensorRT and TritonNeMo, Retriever and GuardrailsNIMH100 or HGX trainingA100 or L40S inference
Planned
DGX central training over multi-site telemetry · plannedFederated multi-site model sharing · plannedActive learning on coupons and destructive tests · planned
Why it needs a GPU
Fleet-scale prediction over multimodal telemetry, cuOpt combinatorial scheduling, RAPIDS analytics on large genealogy tables and standards reasoning are GPU-bound. CPU-only cannot re-optimise the floor or serve reasoning at production cadence.
MVP scope and first metric
Aggregate feeds from Meltrix and Poroscan into a fleet view with build-failure prediction and genealogy. Add cuOpt scheduling and standards-aware qualification reasoning, then graduate to closed-loop orchestration of all four products.
Tier. Core of the Floor tier ($70,000 per month) and the anchor of Enterprise contracts ($500k to $7M ACV).
Open questions
- AM-MES and ERP integration surfaces and data governance vary by customer
- Federated multi-site model sharing without IP leakage is planned
Compare
What is in each tier
Products are licensed per machine on the Machine tier and bundled on Floor and Enterprise.
| Capability | Machine | Floor | Enterprise |
|---|---|---|---|
| Machines covered | 1 | All on the floor | All in group |
| Agent modules | 1 of 5 products | All agents | All agents + custom models |
| Part twin (Parttwin) | No | Included | Custom alloy and polymer models |
| Printer-edge node | Included | Per machine | Fleet managed |
| Bounded parameter writes | Inside the envelope, scoped tags | All scoped tags | Group policy |
| Build genealogy and evidence packets | Included | Included | Included + export and change control |
| Autonomy levels | L1 to L2 | L1 to L3 | L1 to L4 when validated |
| SSO and RBAC | No | Planned | Planned, with approval chains |
| On-prem deployment | No | Optional | Included |
| Support | Business hours | 24/7 | 24/7 + named engineer |
| Indicative price | $6,000/mo | $70,000/mo | $500k to $7M ACV |
Run timeline · scenario
The five products in one build
Every Powdrix run is an ordered, inspectable sequence. This is run_7c2e4b on PBF-2 · Ti-6Al-4V bracket family, 24-part build. Scenario completes in 10 steps with one approval gate. Twin recipe #14 selected, one lack-of-fusion precursor at layer 1,872 cleared by a bounded power trim, CT triage in 11 minutes, 23 of 24 parts released with full genealogy and one held for review.
run_7c2e4b
PBF-2 · Ti-6Al-4V bracket family, 24-part build
illustrative scenario
SUCCEEDED
-
01
ingest.job
SUCCEEDED
0.7 s
Floor Orchestrator · Pulled the job from AM MES: 24 Ti-6Al-4V brackets, 60 µm layers, 3,140 layers, density target 99.7%, geometry tolerance ±0.10 mm, AS9100 evidence packet required. Locked the OEM-approved parameter envelope for PBF-2.
-
02
twin.simulate
SUCCEEDED
6 m 12 s
Part Twin · Simulated 40 candidate orientation, support and parameter recipes for thermal field, residual stress and distortion, ranked on predicted density, geometry deviation, takt and energy.
-
03
prep.release
SUCCEEDED
48 s
Prep-and-Parameter · Recipe #14 released: 32° orientation, lattice supports on two faces, laser power and speed inside the qualified envelope. Predicted distortion 0.06 mm, predicted density 99.8%.
-
04
recoat.check
SUCCEEDED
continuous
Recoat-and-Powder · Recoater image checked every layer. Powder lot L-2291 on its third reuse, humidity in window. One short-feed at layer 412 corrected with a dosing increase before the next layer fused.
-
05
melt.stabilise
SUCCEEDED
continuous
Melt-and-Layer · Melt-pool, photodiode and thermal streams fused at 120 fps. Lack-of-fusion risk rose to 0.61 on part 17 at layer 1,872 and was cleared by a 3% power trim inside the envelope within four layers.
-
06
thermal.compensate
SUCCEEDED
2 m 04 s
Thermal-and-Distortion · Interlayer temperature on the tall brackets ran 14 °C above the twin's prediction from layer 2,300. A 1.5 s interlayer dwell held the thermal field inside the predicted band through the top section.
-
07
approve.human
APPROVAL
4 m 30 s
Floor Orchestrator · Melt-and-Layer proposed a hatch-spacing change on part 17 beyond the qualified envelope. Held for the AM process engineer, who declined it and chose to flag part 17 for CT priority instead. Decision logged.
-
08
handle.depowder
SUCCEEDED
38 m
Robot-and-Handling · Plate unloaded by the handling cell on build complete, depowdered, and parts transferred to stress relief. Powder returned to the sieve with lot L-2291 reuse count incremented in genealogy.
-
09
inspect.ct
SUCCEEDED
11 m
Defect-and-Inspect · 24 CT volumes segmented for pores, cracks and lack of fusion in 11 minutes. 23 parts graded accept against the acceptance criteria. Part 17 shows a 180 µm indication at layer 1,870 and is routed to the inspector for disposition.
-
10
release.qualify
SUCCEEDED
1 m 20 s
Quality-and-Conformance · 23 parts released with the full genealogy: recipe, layer tags, melt-pool events, thermal trace, approvals, post-processing cycles and CT result. Part 17 held with a non-conformance opened. Evidence packet assembled for sign-off.
Accelerated computing
Physical AI, at the melt pool
A build is thousands of layers and a day of machine time. Perception has to be local, deterministic and fast, CT has to be triaged in minutes, and the twin has to run before the build. Powdrix runs GPU inference at the printer edge and keeps training, simulation and optimisation in the cloud or on-prem.
Edge perception
One Jetson node per printer for 30 to 240 fps melt-pool, layer and recoater streams with photodiode and thermal fusion. Sub-250 ms alerting and 1 to 2 s bounded recommendations are the design targets. Jetson Thor is planned for multi-laser machines.
CT and vision inference
2D and 3D segmentation models for porosity, cracks, lack of fusion, surface defects and metrology comparison, served per inspection cell. Sub-15-minute CT triage per build is the design target.
Training
Defect, melt-pool-instability, build-failure and AM reasoning models trained weekly over multimodal build data, with gated promotion from shadow to assist to bounded. Dedicated capacity is planned.
Part twin
Thermal, melt-pool-proxy, residual-stress and distortion simulation for recipe ranking, plus robot-cell and queue simulation. Modulus surrogates are planned so sweeps run at production cadence.
Synthetic data
Rare pore, lack-of-fusion, spatter, recoat-streak and distortion scenarios generated and validated against real CT and in-situ data before promotion. 1,000 to 10,000 per qualified part family is the initial target.
Optimisation and analytics
Build nesting, fleet scheduling across machines, ovens, HIP, inspection and robots, energy windows, and GPU analytics over genealogy tables.
Robot handling and serving
Plate, powder and part handling policies trained in simulation and run on Jetson in the cell. NIM packages perception, prediction and reasoning services behind the floor orchestrator.
Pricing
Priced per machine, per floor, per group
Land on one printer with a measured baseline. Expand by machine, module, floor and autonomy level. Outcome components are available on build failure, yield, qualification and energy. Annual prepay carries a 15 to 20% discount.
In-situ, parameter-control or inspection AI for one printer. For manufacturers validating ROI on the wedge.
- One printer or inspection cell, one agent module
- Printer-edge Jetson node included
- Machine, camera and CT reads, bounded writes inside the envelope
- Run log, audit trail, build genealogy
- Business-hours support
The whole loop: prep and parameter, melt and layer, recoat and powder, thermal and distortion, defect, post-process, yield and takt, robotic handling and the twin.
- All agents on every machine on the floor
- Poroscan on every inspection cell
- Yieldex scheduling, genealogy and conformance
- Forgebot handling cells
- Parttwin recipe optimisation before every build
- 24/7 support, quarterly model review
Multi-site AM manufacturers and bureaus standardising on Powdrix, with custom alloy, polymer and qualification models.
- Multi-site floor-edge fleet
- Custom alloy, polymer and qualification models
- On-prem GPU inference option
- Group autonomy policy and approval chains
- SLAs, SSO and RBAC, dedicated engineer
- Validated model change control and evidence packs
FAQ
Straight answers
The questions operations directors, process engineers and certification engineers ask in the first meeting.
Only inside the OEM-approved parameter envelope, only on an explicit allow-list with per-tag limits, and only at the autonomy level your floor has set. Level 1 is advisory: Powdrix predicts and recommends and a human enters everything. Design partners start there, in shadow mode. Anything that leaves the envelope or affects qualification always waits for a named approver.
Control returns to the printer's own controller and the standing recipe. Powdrix is a supervisory layer on top of the machine controls, monitoring systems and MES you already run, never a replacement for them, so an outage degrades the floor to its current way of running, not to a stop.
Advisory recommendations start from the twin's physics and the part spec in the first weeks. Defect and build-failure prediction improve as layers, CT results and engineer corrections accumulate per part family. The first models are trained on existing in-situ camera data; synthetic rare-defect scenarios augment them only after validation against real CT.
Only if you choose cloud training. Parameter sets, geometry, defect libraries and process memory are isolated per tenant and never used to train another customer's models. Cross-customer learning is federated and IP-isolated and waits on written data agreements. A fully on-prem deployment is available for sensitive producers.
It is designed not to. Validated parameters never change without the sign-off your quality system requires, model changes go through change control, and every action carries model version, data lineage, approval status and qualification impact in the audit log. Powdrix does not claim AS9100 or ISO 13485 certification; assurance packages are planned for the 6 to 12 month release.
You are, as with any process change, which is why every write is policy-checked, bounded, logged and reversible, and why anything that affects qualification waits for a named approver. The audit log records the request, the reasoning, the limits applied and the human decision.
Get started
Pick your first product
Start with one printer or one inspection cell and one measurable baseline. A paid pilot on melt-pool and build-failure prediction, CT porosity triage or pre-build recipe optimisation shows the number before you commit further.