Rivetira
Book a line assessment

Everything needed to run agents on a real assembly line

Written for manufacturing engineers, controls engineers and quality engineers — not for a generic developer audience. Every page assumes you have a station, a structure and a takt to hold.

Documented agents

7

all production agents

Integration guides

24

machine and system classes

Median time to first query

11 min

from install

Docs updated

Weekly

with each release

Overview

Rivetira runs as a factory edge runtime inside your plant network. It reads from the sensors, metrology systems and machine controllers already present at a station, builds an as-built model of the structure in front of it, and — at the autonomy level you have granted — writes instructions back to the machines.

Nothing about that loop depends on a cloud connection. The control plane exists to deliver models, aggregate fleet analytics and support your engineers. If it disappears, the line keeps running.

The line assessment

Every deployment starts with a three-week assessment on one station. We instrument the station, record the current process, and quantify the rework, shim and takt reality against which everything afterwards is measured. The baseline that comes out of this is signed by both parties before an agent runs.

Installing the edge runtime

The runtime ships as a signed OCI image plus a hardware profile for the station class. It requires no inbound network ports. Installation on a supported industrial PC takes under an hour; accreditation on an air-gapped program takes considerably longer and is scheduled with your security team.

Your first station

A new station always starts in shadow mode. Agents observe, predict and record, but write nothing. You compare their predictions against what your mechanics actually did. Only when measured accuracy clears the promotion gate does advisory mode become available.

Autonomy is never granted by Rivetira. It is granted by your site engineering, per station, per agent, and can be revoked in a single action.

The four levels, and what each one permits

Autonomy levels and permitted actions
LevelAgent mayHuman mustPromotion gate
ShadowObserve, predict, recordPerform every operationNone — entry level
AdvisoryRecommend a specific actionAccept or reject each recommendation≥ 95% prediction accuracy over 4 weeks
SupervisedExecute with a human present and armedArm the operation, remain at the station≥ 98% accuracy, twin agreement, 0 safety events
Closed loopExecute within the qualified envelopeReview exceptions and periodic audits≥ 99% accuracy over 8 weeks, site sign-off
87.5%90.0%92.5%95.0%97.5%100.0%Agreement with physical outcomeWk 2Wk 6Wk 10Wk 14Wk 18Wk 22Wk 26Wk 30spec limitMeasured accuracyTwin agreement
A typical station climbing the autonomy gatesThe value axis is truncated to resolve the gates; full range in the data table. The 99% line is the closed-loop gate.
A typical station climbing the autonomy gates
Week in deploymentMeasured accuracyTwin agreement
Wk 291.2%88.4%
Wk 694.1%92.0%
Wk 1096.3%94.8%
Wk 1497.4%96.6%
Wk 1898.2%97.9%
Wk 2298.7%98.5%
Wk 2699.1%99.0%
Wk 3099.4%99.3%

Perceive, plan, control, verify, learn

The same loop runs at every station, from a single drilling cell to a whole moving line. Nothing is actuated that has not first been simulated against the as-built twin.

  1. 01

    Perceive

    Vision, in-process metrology, laser tracker and photogrammetry fuse into a live as-built model of the structure, hole, gap and fastener state.

    38 msFusion latency p99

  2. 02

    Plan

    Align, shim, drill, countersink, fasten, seal and join are sequenced against the predicted gap field and the program tolerance stack-up.

    4.1 sTwin sim per join

  3. 03

    Control

    Feed and speed, interference, torque, bead geometry and crawler position are driven adaptively, with fail-safe stop at every actuator.

    240 HzControl loop

  4. 04

    Verify & learn

    Hole, countersink, fastener, gap and FOD are sensed, conformance is logged immutably, and supervised corrections retrain the models.

    1.4MTraced actions/day

As-built twin gates every step Perceive 38 ms fusion Plan 4.1 s twin sim Control 240 Hz loop Verify & learn 1.4M traces/day

A station is a file

stations/WG-11.yaml
station: WG-11
description: Wing skin-to-spar, lower skin
structure: narrowbody-wing-box
material_stack: [CFRP-6.4mm, Ti-6Al-4V-3.2mm]

agents:
  drilling_fastening:
    autonomy: closed_loop
    envelope:
      feed_mm_per_rev: [0.04, 0.11]
      spindle_rpm: [2200, 5400]
      thrust_n_max: 980
    on_envelope_violation: reject_and_alert

  metrology_shimming:
    autonomy: closed_loop
    gap_prediction_mae_mm_max: 0.045
    shim_machine: haas-5ax-02

  inspection:
    autonomy: supervised
    coverage: all_holes
    escalate_severity_at: 3

acceptance:
  requires_human_signature: true
  blocking_conditions: [fod_detected, twin_drift]

Envelope enforcement

The envelope block is the qualified process window for that material stack. It is enforced at the edge, in the control path, before any instruction reaches a machine. An agent that plans an operation outside the envelope has its instruction rejected and an alert raised — regardless of how confident the model is.

Rollback

Changing autonomy: closed_loop to advisory and applying takes under 60 seconds and is a logged, attributed event. Any authorised site engineer can do it. Nothing about rollback requires Rivetira to be reachable.

Blocking conditions halt station acceptance. A serial with an open FOD detection or unresolved twin drift cannot be signed off, by anyone, until the condition is dispositioned.

Hit the fit before the join

Every airframe diverges from CAD the moment the first hole is drilled. The as-built twin models that divergence — fit, gap, shim, drilling and structural conformance — and simulates the join before a single fastener is installed.

  • Simulates fit, gap field and shim geometry against the real as-built structure, not the as-designed model
  • Gates every autonomy promotion: an agent only takes a step in the plant after it takes it in the twin
  • Predicts tolerance stack-up across the whole join sequence, not point by point
  • Replays any historical join for root-cause analysis and airworthiness investigation
0.0 mm0.2 mm0.4 mm0.6 mmPeak gap (mm)Section 41Section 43Section 44Wing box LWing box REmpennage
Twin-predicted vs measured peak gap, six major joinsMean absolute error 0.031 mm across 1,900 modelled joins.
Twin-predicted vs measured peak gap, six major joins
StructureTwin predictionMeasured as-built
Section 410.31 mm0.34 mm
Section 430.44 mm0.41 mm
Section 440.28 mm0.30 mm
Wing box L0.52 mm0.55 mm
Wing box R0.49 mm0.47 mm
Empennage0.22 mm0.24 mm

Connecting the systems you already run

Integration mechanisms by system class
System classMechanismDirectionTypical effort
Assembly MESREST + webhooksBidirectional2–4 weeks
PLM / as-designed modelsSTEP AP242 + JT importRead1–2 weeks
ERP work ordersRESTRead1 week
Quality / NCR systemsWebhooks + REST write-backBidirectional2–3 weeks
Machine controllersOPC UA / MTConnectBidirectionalPer machine class
Laser trackers & scannersVendor SDK adaptersReadIncluded
Shim machining cellsG-code + OPC UAWrite1–2 weeks
HistoriansOPC UA HDAWrite1 week

Running the edge fleet

bash — fleet operations
$ rivetira fleet status --site everett
RUNTIME   STATION  VERSION   AUTONOMY     HEALTH   UPTIME
edge-01   FA-01    2026.6.2  closed-loop  ok       61d 04h
edge-02   FA-02    2026.6.2  supervised   ok       61d 04h
edge-03   WG-11    2026.6.2  closed-loop  ok       48d 11h
edge-04   WG-12    2026.6.1  advisory     update   14d 02h

$ rivetira fleet upgrade edge-04 --to 2026.6.2 --window next-shift-change
✓ Artefact signature verified
✓ Scheduled for 2026-08-14 06:00 local (shift change)
✓ Station will drop to advisory during upgrade — line unaffected

Observability

Every runtime exports metrics in Prometheus format and traces in OpenTelemetry, into your own observability stack. There is no requirement to send anything to Rivetira to know whether your own line is healthy.

Backup & restore

As-built records replicate to your storage continuously with a 15-minute RPO and a 4-hour RTO. Restore is tested quarterly as part of the service, and the hash chain is verified end-to-end on every restore.

Incident response

A runtime that cannot verify its own model artefacts refuses to leave advisory mode. A runtime that loses metrology input stops planning rather than planning blind. Both conditions page your on-call and Rivetira's simultaneously on connected deployments.

Built for programs that cannot leak and cannot fail

Airframe geometry is among the most protected IP in manufacturing, and much of it is export-controlled. Rivetira is architected for that reality from the edge up.

Certified

SOC 2 Type II

Annual audit covering security, availability and confidentiality of the control plane.

Supported

ITAR / EAR aware

US-person access controls, on-prem and air-gapped deployment for controlled programs.

Mapped

AS9100 alignment

Quality records, NCR flow and traceability mapped to aerospace quality requirements.

In progress

ISO 27001

Information security management system certification underway. [PLACEHOLDER: target date]

Data stays where the airframe is

  • Factory edge runtime holds all program geometry, imagery and metrology locally
  • Air-gapped mode: model updates arrive by signed media, telemetry never leaves
  • Per-tenant isolation and per-program cryptographic separation
  • Encryption in transit and at rest; customer-managed keys available

Assurance-grade by default

  • Immutable audit trail for every agent action, sensor reading and human override
  • SSO/SAML, SCIM provisioning and role-based access down to station level
  • Fail-safe stop for drills, fasteners, crawlers and cranes at < 40 ms
  • Human-in-the-loop mandatory for airworthiness-critical dispositions

Reads and writes back to the machines you already own

Rivetira is not a rip-and-replace. It connects to the drilling machines, crawlers, trackers, shim cells, sealant robots and MES already on your floor over APIs and OT protocols.

Drilling & fastening

  • Flex-track crawlers
  • Gantry drilling machines
  • C-frame riveters
  • Automated fastening cells
  • Hand-fed ADUs

Metrology

  • Laser trackers
  • Photogrammetry rigs
  • Structured-light scanners
  • iGPS / indoor GPS
  • Articulated arm CMMs

Shimming & sealing

  • 5-axis shim machining
  • Liquid shim dispense
  • Sealant robots
  • Cure monitoring
  • Bead inspection

Factory systems

  • Assembly MES
  • PLM / as-designed models
  • ERP work orders
  • Quality / NCR systems
  • Historians & OPC UA

OPC UA, MTConnect, MQTT, ROS 2, REST and file-drop connectors. Read-only shadow mode first; write-back enabled only after twin validation.

Every station, every structure, one record

Sort any column. The same table backs the audit export an airworthiness engineer hands to a regulator.

Assembly station performance, trailing 30 days
FA-01 · Fuselage section joinSection 41/4399.2%0.041 mm18 h96.4%
FA-02 · Wing-body joinWing box98.6%0.062 mm31 h94.1%
FA-03 · Empennage attachVertical/horizontal99.5%0.028 mm9 h98.2%
FA-04 · Systems installationHydraulics/electrical97.8%0.055 mm12 h95.8%
FA-05 · Final assemblyInteriors/doors98.9%0.037 mm6 h96.9%
WG-11 · Wing skin-to-sparLower skin99.1%0.033 mm22 h97.4%
WG-12 · Wing skin-to-sparUpper skin98.4%0.048 mm26 h95.2%
FS-21 · Fuselage panelBarrel section99.3%0.030 mm11 h97.9%

Questions engineers actually ask

Anything else goes to the people who build it. Ask an assembly engineer or read the full FAQ.

The conceptual documentation is public — what you are reading. Program-specific integration guides, the envelope qualification methodology and the API reference are available to customers and to prospects under NDA.

Weekly, with each release. Breaking changes to configuration schema carry a twelve-month deprecation window and are announced in the release notes before they ship.

Yes. Factory and Enterprise deployments include a three-day on-site course for manufacturing, controls and quality engineers, plus a shorter operator module delivered at the station.

Get the full documentation

A line assessment maps one station, quantifies the rework, shim and rate opportunity, and returns a modelled ROI in three weeks. No production disruption.