Developers

Typed tools. Declared limits. Replayable runs.

The Hullira SDK is a design preview of how an agent is defined, bounded and run: tools with unit-aware ranges, rate limits and WPS, coating and dimensional envelopes, autonomy policy as code, deterministic replay of any historical run against a new model, and a local twin harness. The API will change before release.

Python SDK · design previewCLI · design previewPolicy as code

run_4b7e19 · agent graph · block B-214 · scenario 10 nodes · 1 approval gate
01 · orchestrator ingest.package ✓ SUCCEEDED 02 · aligna twin.simulate ✓ SUCCEEDED 03 · cut_profile cut.nest ✓ SUCCEEDED 04 · panel_weld weld.sequence ✓ SUCCEEDED 05 · orchestrator approve.human ◆ APPROVAL 06 · quality_anomaly inspect.fuse ✓ SUCCEEDED 07 · fair_form fair.lineheat ✓ SUCCEEDED 08 · blast_coat coat.window ✓ SUCCEEDED 09 · schedule_yard dock.reoptimise ✓ SUCCEEDED 10 · erect_align erect.align ✓ SUCCEEDED

Illustrative scenario: Block B-214 · double-bottom block · panel line PL-2 · dock D1, one block, one approval gate

Concepts

Six objects

Everything in the SDK is one of these.

Agent

A goal, tools and an autonomy level

An agent declares what it is for, what it may call and how far it may go without a human. Its runs are typed and logged.

Agent

Tool

A connector call with a schema

Tools name tags, units and whether they are read-only. Write tools carry Limits and an envelope.

Toolread_only

Limit

Rate, magnitude, envelope

max_step per period, absolute floors and ceilings, and the qualified WPS, coating or dimensional envelope, enforced at call time by the policy engine and again at the OT boundary.

Limitenvelope

Autonomy

L1 to L4

The level the yard has set for this agent, this tag, this shift. Approvals are routed to named humans above the threshold.

Autonomy.L1Autonomy.L3

Run

A goal becoming a graph of steps

Start, stream, replay. Every step has an agent, a tool, a duration and a status, and the whole record is kept per block.

run.stream()run.replay()

Twin

verify=twin

Simulate a move on the as-built ship twin before it is written. Required for any agent configured with it.

twin.simulate
Developers

Define an agent, bound it, run it

The Hullira SDK is typed Python. Tools are declared with schemas, limits and the qualified WPS envelope, and the policy engine enforces them at call time, not in a review meeting. This is a design preview; the API will change before release.

What you get

  • Typed tool definitions with unit-aware ranges, rate limits and WPS, coating and dimensional envelopes
  • Deterministic replay of any historical run against a new model
  • Local twin harness so a weld sequence is simulated before it is shipped
  • Autonomy policy as code, versioned and reviewed like any other change

Read the docs Developer guide

pl2_weld_agent.py
# Design preview. Bound the weld agent to robot cell R3 on panel line PL-2.
from hullira import Agent, Tool, Limit, Autonomy

params = Tool(
    name="robot.set_parameters",
    tags=["PL2.R3.WIRE_FEED", "PL2.R3.TRAVEL_SPEED"],
    limits=[Limit(max_step="4%", per="60s", envelope="WPS-117")],
)

weld = Agent(
    id="agent.panel_weld",
    goal="porosity risk < 0.1 per joint, inside WPS-117, least rework",
    tools=[params, Tool("weld.read_pool", read_only=True)],
    # bounded writes on parameters; class-critical seam order stays with a human
    autonomy=Autonomy.L3,
    # simulate on the twin before every write
    verify="twin",
)

run = weld.start(line="PL-2", block="B-214")
for step in run.stream():
    print(step.name, step.status, step.duration)
CLI

Start a run from anywhere

The same run engine, the same policy checks, the same audit trail, from the terminal, the review console or the SDK. Design preview.

hullira · cli · scenario
$ hullira run "block B-214 on PL-2 to D1" --autonomy L2+

→ plan composed          10 steps · 1 approval gate
→ twin.simulate          18 candidates · best #6 · joint 1.8 mm
→ policy.evaluate        8 writes permitted · class-critical seams held for human
→ executing              cut.nest .......... ok   3m40s
→ executing              weld.sequence ..... ok   2m05s
! approval required      seam_sequence S-31, S-32
→ approved               shift welding engineer · 06:12:44
→ executing              inspect.fuse ...... ok   continuous
→ executing              fair.lineheat ..... ok   4m30s
→ scenario complete      landed 1.6 mm · conformance record written

$ hullira runs show run_4b7e19 --format genealogy
Run timeline · scenario

One scenario run, step by step

Every Hullira run is an ordered, inspectable sequence. This is run_4b7e19 on Block B-214 · double-bottom block · panel line PL-2 · dock D1. Scenario completes in 10 steps with one approval gate. Predicted distortion at the erection joint fell from 4.2 mm to 1.8 mm after a back-step sequence, one porosity precursor on seam S-31 was cleared inside the WPS envelope, the ballast-tank coat ran inside a 3.4 °C dew-point margin, and the block landed at 1.6 mm with its conformance record written.

run_4b7e19 Block B-214 · double-bottom block · panel line PL-2 · dock D1 illustrative scenario SUCCEEDED
  1. 01 ingest.package SUCCEEDED 0.8 s

    Yard Orchestrator · Pulled the work package for B-214 from the yard MES: 46 plates, 212 seams under WPS-117, erection tolerance 2 mm at the D1 joint, PSPC-grade ballast-tank coating, erection slot on D1 in the dock plan. Locked the qualified WPS envelope and the class scope.

  2. 02 twin.simulate SUCCEEDED 1 m 12 s

    Ship Twin · Simulated 18 candidate weld sequences on the as-built twin with the weld-thermal surrogate, ranked on predicted distortion at the erection joint, crane reach on D1 and shift takt. Standing sequence predicts 4.2 mm; candidate #6 predicts 1.8 mm.

  3. 03 cut.nest SUCCEEDED 3 m 40 s

    Cut-and-Profile · Released the nest for 46 plates on CL-1 with the fit-up gaps candidate #6 assumes. As-cut geometry logged per plate and handed to the panel line; material yield 91.4% against the nest plan.

  4. 04 weld.sequence SUCCEEDED 2 m 05 s

    Panel-and-Weld · Loaded the back-step sequence for the non-critical seams on PL-2 and set heat input and travel speed inside WPS-117. Weld-pool and seam cameras streaming on 12 channels.

  5. 05 approve.human APPROVAL 58 s

    Yard Orchestrator · Candidate #6 also re-orders two class-critical butt seams, S-31 and S-32. Sequence changes on class-critical seams stay with a human, so the step was held for the shift welding engineer and approved at 06:12:44.

  6. 06 inspect.fuse SUCCEEDED continuous

    Quality-and-Anomaly · Twelve weld cameras and the UT station fused at the yard edge. One porosity precursor on S-31 at 06:41 flagged and cleared by a 4% wire-feed trim inside the WPS-117 envelope; the joint was routed to UT and passed.

  7. 07 fair.lineheat SUCCEEDED 4 m 30 s

    Fair-and-Form · Post-weld scan registered against the twin: 2.1 mm at the erection joint. Two line-heating passes planned on the inner bottom and executed by the fairing crew; re-scan read 1.8 mm, inside the 2 mm tolerance.

  8. 08 coat.window SUCCEEDED continuous

    Blast-and-Coat · Ballast-tank blast and first coat scheduled to a window with steel temperature 3.4 °C above dew point. Dry-film thickness read 312 µm against a 280 µm minimum on the stripe coats; PSPC evidence assembled per area.

  9. 09 dock.reoptimise SUCCEEDED 41 s

    Schedule-and-Yard · Re-solved the D1 dock and crane plan with B-214 ready 6 hours early and a neighbouring block late: erection slot pulled forward one shift, crane G2 reassigned, no berth conflict. Planner approved the revised plan.

  10. 10 erect.align SUCCEEDED 1 h 48 m

    Erect-and-Align · Both blocks scanned on D1 and registered against the twin; fit predicted at 1.6 mm. Shimming guidance issued to the erection crew, block landed and welded out, final scan 1.6 mm. Conformance record written: welds, scans, coating readings, approvals and class evidence for B-214.

Tool calls · scenario

The agent shows its work

Every tool invocation, argument and result is written to an append-only, human-readable log, and every reasoning step is expandable. Nothing about a run is hidden from the yard or the surveyor.

tool-call stream · run_4b7e19
  1. 06:04:10mes.read_packages({ block: "B-214", dock: "D1" })
  2. 06:04:11ok 46 plates · 212 seams · WPS-117 · 2 mm at D1 joint
  3. 06:04:12twin.simulate({ candidates: 18, objective: "distortion, reach, takt" })
  4. 06:05:24ok best #6 · joint 1.8 mm · reach ok · takt +0.4 h
  5. 06:05:25policy.evaluate({ plan: 6, autonomy: "L2+" })
  6. 06:05:25ok 8 writes permitted · class-critical seam order requires human approval
  7. 06:05:26cutter.set_program({ machine: "CL-1", nest: "B214-N3" })
  8. 06:09:06ok 46 plates cut · yield 91.4% · geometry logged
  9. 06:09:07weld.set_sequence({ line: "PL-2", plan: 6, scope: "non-critical" })
  10. 06:11:12ok back-step loaded · heat input inside WPS-117
  11. 06:11:46approval.request({ action: "seam_sequence", seams: ["S-31", "S-32"] })
  12. 06:12:44ok approved by shift welding engineer · logged
  13. 06:12:45sensors.fuse({ cams: 12, ut: true })
  14. 06:41:03precursor: porosity risk 0.57 · seam S-31 · 1.2 m from start
  15. 06:41:04robot.set_parameters({ cell: "PL-2.R3", wire_feed: "-4%", within: "WPS-117" })
  16. 06:41:19ok porosity risk 0.57 → 0.09 · S-31 routed to UT · passed
  17. 09:20:30scan.register({ block: "B-214", stage: "post-weld" })
  18. 09:21:02ok joint 2.1 mm · 2 line-heat passes planned
  19. 11:02:40ok re-scan 1.8 mm · inside 2 mm tolerance
  20. 11:02:41coat.schedule_area({ area: "BT-214", spec: "PSPC" })
  21. 11:02:42ok window open · steel 3.4 °C above dew point · DFT 312 µm
  22. 11:03:00cuopt.solve({ dock: "D1", horizon: "14d" })
  23. 11:03:41ok slot pulled forward 1 shift · crane G2 reassigned · planner approved
  24. 14:30:12align.guide({ block: "B-214", dock: "D1" })
  25. 16:18:40ok landed 1.6 mm · conformance record written · scenario complete

Reasoning trace

Plan, thought, action, observation, exactly as the orchestrator would record it.

GoalBuild and erect B-214 inside WPS-117, hold the erection joint …

Build and erect B-214 inside WPS-117, hold the erection joint within 2 mm, coat the ballast tank inside the PSPC window, and land it on D1 by the dock plan.

PlanSimulate weld sequences on the twin first. Lead with the seque…

Simulate weld sequences on the twin first. Lead with the sequence because distortion is decided at the arc, then cut to the fit-up the sequence assumes, then weld and inspect continuously. Scan after welding and fair only what the scan shows. Hold any class-critical seam change for a human.

ThoughtThe standing sequence welds the inner bottom straight through …

The standing sequence welds the inner bottom straight through and predicts 4.2 mm at the joint, which means fairing hours and a risk to the D1 slot. A back-step sequence with two class-critical seams re-ordered predicts 1.8 mm at a cost of 0.4 hours of takt.

Actiontwin.simulate(candidates=18) returns plan #6: back-step on the…

twin.simulate(candidates=18) returns plan #6: back-step on the non-critical seams, S-31 and S-32 re-ordered, heat input held inside WPS-117.

ObservationPlan #6 predicts 1.8 mm at the joint and clear crane reach on …

Plan #6 predicts 1.8 mm at the joint and clear crane reach on D1. Three candidates scored lower on distortion but moved heat input outside the WPS envelope and were discarded.

ActionExecute plan #6: eight bounded writes permitted on cutting, no…

Execute plan #6: eight bounded writes permitted on cutting, non-critical weld sequence, parameters inside WPS, coating windows and dock plan; the class-critical seam order routed to the shift welding engineer.

ObservationPorosity precursor on S-31 at 06:41:03. A 4% wire-feed trim in…

Porosity precursor on S-31 at 06:41:03. A 4% wire-feed trim inside WPS-117 cleared it in 15 s; the joint was routed to UT and passed. Post-weld scan 2.1 mm, two line-heating passes brought it to 1.8 mm.

OutcomeScenario complete. Block landed at 1.6 mm, ballast tank coated…

Scenario complete. Block landed at 1.6 mm, ballast tank coated inside the window, conformance record written for class review.

Autonomy

Four levels, set per agent and per tag

A yard does not go from paper travelers to unattended welding in one step. Hullira makes the level explicit, auditable and reversible at any time. The first release runs shadow mode, then assist mode, then supervised control on low-risk actions. Class-critical seam changes, lift plans and block release stay with named people.

Autonomy levels and the human role at each
LevelWhat the agent doesWhat the human doesWhere it sits in a pilot
L1 · ShadowReads weld, scan, NDT, coating, robot, crane and MES signals; predicts and recommends; writes nothingRuns the process as today; compares predictions to NDT and surveyor outcomesShadow mode; the baseline is measured from the yard's own records
L2 · AssistProposes weld, fairing, coating, routing and schedule moves with evidence, confidence, envelope and rollback pathApproves each action in the review consoleAssist mode, the remainder of the pilot
L3 · Supervised controlExecutes pre-approved low-risk actions: inspection routing, coating pauses when a window drifts, weld parameters inside the WPS envelope, robot path variants tested in simulationApproves class-critical seam changes, lift plans and block releaseGraduated autonomy, after audited accuracy and ROI
L4 · BoundedSupervised work sequencing and weld adaptation for qualified block types, across processes and yardsSets the envelope; reviews the shift record and the evaluation resultsPlanned; after class, yard and insurer acceptance is earned
Accelerated computing

Physical AI, at the arc

A weld pool is decided in milliseconds and a dock waits by the hour. Perception has to be local, deterministic and fast, so Hullira runs GPU inference at the yard edge and keeps training, simulation and optimisation in the cloud or on-prem.

Jetson Orin · DeepStream · Holoscan · TensorRT

Edge perception

6 to 24 cameras per process area plus UT, X-ray, thermal, laser-scan, photogrammetry, robot and coating-instrument streams fused at the yard edge. Sub-50 ms weld alerts and sub-150 ms coating or scan prompts are the design targets. Jetson Thor is planned for higher camera density.

Triton · NIM

Model serving

Weld-defect, distortion-risk, coating-risk, point-cloud and schedule-risk models behind one orchestrator at the yard edge, versioned with rollback, routed by latency, tenant, sensitivity and cost.

DGX or HGX · NeMo

Training

Weld-defect vision, UT and X-ray classification, NeMo class-rule reasoning and weld-thermal FEA surrogates fine-tuned per tenant: 100 to 500 GPU-hours per yard during onboarding, then monthly retraining from engineer corrections and class outcomes. Dedicated training capacity is planned.

Omniverse · OVX · Modulus

Ship and yard twin

Block erection, weld sequence, robot reach, crane paths and yard layout simulated before the shift, as-built against as-designed. GPU weld-thermal and structural FEA is the workload; physics-informed surrogates rank plans at shift cadence.

Cosmos · Omniverse Replicator

Synthetic data

Rare weld defects, coating failures, distortion modes, occluded seams and erection misalignment cases generated and validated against real NDT and scan distributions before promotion. 50k to 250k variants per vessel class is the target. Planned.

cuOpt

Optimisation

Dock, crane and berth schedules, work-package sequencing, yard logistics and erection order, re-solved as progress changes.

Isaac · Isaac ROS

Robot cells and crawlers

Robotic weld, blast, coat and inspection cells and crawler or drone hull inspection, with paths validated in simulation before they run next to a hull.

RAPIDS

GPU ETL

Telemetry ETL across weld parameters, scans, NDT results, coating readings, non-conformance and schedule history, for retraining, drift detection and rework, dock-time and schedule benchmarks.

NeMo Guardrails · Retriever

Policy and grounding

WPS and PQR, class-rule, PSPC, cyber and human-approval policies enforced before any action writes back to robots, MES, QA or work packages; retrieval with citations and tenant isolation.

Guardrails

An agent that can move a weld parameter needs a leash

Hullira writes to production equipment. Every capability is scoped, every write is policy-checked against the qualified WPS, class and PSPC constraints, and every action is written to an append-only audit log the yard owns.

  • Bounded action space. Each agent writes only to an explicit tag allow-list, inside per-tag rate and magnitude limits and the qualified WPS, coating, dimensional and schedule constraints for the block.
  • Policy engine before every write. WPS and PQR, class rules, PSPC, safety limits, role permissions, autonomy level and human-approval requirements are evaluated before a parameter moves. NeMo Guardrails is the design for this layer.
  • Human-in-the-loop gates. Class-critical seam changes, lift plans, block release and class evidence wait for a named approver. Robots adapt only inside pre-approved WPS envelopes until class, yard and insurer acceptance supports more.
  • Immutable audit log. Append-only, assurance-grade and exportable. Every action records source evidence, the WPS, class or PSPC citation applied, confidence, estimated rework, dock-time and schedule impact, rollback path, approver and block scope.
  • Hard fallback. Loss of the edge node, the network or a model returns control to the robot controller, PLC or planner's last known-good state. Interlocks are never in Hullira's path.
  • Tenant and defense-aware isolation. Drawings, WPS, naval data, models and memory never cross a customer boundary. Security is defense and ITAR-aware, and on-prem or yard-edge deployment is available for naval and regulated yards.

Compliance posture

Compliance and certification status
ItemScopeStatus
SOC 2 Type ICloud control plane QUEUED Planned, first 6 months, with a defense and ITAR-aware review
SOC 2 Type IICloud control plane QUEUED Planned, 6 to 12 months
SSO and RBACConsole and approvals QUEUED Planned, 6 to 12 months
Class-society and defense assurance packagesDisposition and genealogy evidence QUEUED Planned, 6 to 12 months; not a class approval or certification claim
Cited reasoningWPS and PQR, ABS, DNV, LR, ClassNK, IMO PSPC, naval specs RUNNING In the design, every answer cites its source
Per-tenant and IP isolationDrawings, WPS, naval data, models, memory RUNNING In the design from day one
On-prem and yard-edge deploymentNaval and regulated yards RUNNING Design option

Read the security overview

FAQ

Straight answers

The questions production directors, welding engineers and class-compliance leads ask in the first meeting.

Yes, but only within an explicit tag allow-list with per-tag rate and magnitude limits, inside the qualified WPS, coating, dimensional and schedule constraints, and only at the autonomy level your yard has set. Design partners start in shadow mode, where Hullira writes nothing. Class-critical seam changes, lift plans and block release stay with named people.

Get started

Ask for early access

The SDK and CLI are design previews. Tell us what you would build and we will tell you where it stands.