Integrator-led deployment

From Site Survey to First Work-Order Outcome

Five stages, week-by-week. Real Sparkplug topics, real CADET-formatted alerts, real customer takeover. No lorem ipsum.

Week 0 — Site surveyWeek 1 — Hardware installWeek 1-2 — First data pointWeek 2-3 — First action loopWeek 4+ — Customer takeover
1

Week 0

Site survey + scoping

1 day on-site

Your integrator visits the plant, documents the existing systems (PLCs, drives, historian, MES), and maps each production line to the ISA-95 hierarchy. The output is a scoping document that pins every machine to a Site → Area → Cell → Equipment path. Hardware-agnostic — any Docker-capable edge compute works; standard protocols are pre-supported.

Common questions

Do we need to expose our network to the internet?

No. The edge gateway is cellular-first. Nothing routes through the customer's IT firewall by default. Optional Tailscale mesh VPN if remote support is needed.

What if our equipment isn't on the supported list?

We support Sparkplug B, MQTT, OPC UA, Modbus TCP/RTU, EtherNet/IP, MTConnect, and LoRaWAN out of the box — covers ~95% of installed manufacturing equipment. For unusual protocols, the integrator builds a small Node-RED bridge.

How much does the survey cost?

The integrator typically charges a flat survey fee. Flowstate doesn't charge a setup fee — pricing is purely per-machine subscription.

Scoped ISA-95 hierarchy

Acme_Foods
  Massillon_Plant
    Bottling_Area
      Filler_Line_1
        Filler_Krones_01
        Capper_Crown_01
        Labeler_KHS_01

Protocols inventoried

OK Krones Filler — OPC UA (port 4840)
OK Crown Capper — Modbus TCP (port 502)
OK KHS Labeler — EtherNet/IP (port 44818)
WARN Legacy mixer — RS-485 Modbus RTU (gateway needed)

Output: scoping doc + bill-of-materials handed to the customer for internal sign-off.

Stage 1 complete — ready for stage 2
2

Week 1

Hardware install

One shift on-site

Your integrator installs the V4.2 site hub (OnLogic CL260 in a NEMA-4 enclosure) plus any cell-edge nodes near machines that need local store-and-forward. Cellular LTE for WAN, LoRaWAN gateway for wireless sensors, WiFi 6E AP for in-plant uplink. All powered from a single 24V DC bus with supercap backup. Hardware-agnostic — any Docker-capable x86 edge compute will run the same stack.

Common questions

Does the hub touch our PLCs?

Read-only by default. The hub subscribes to data via the protocol stack. Write paths require explicit operator authority gates and an audit log entry per command.

What's the network footprint?

10.0.100.0/24 air-gapped OT network behind the hub. Only the LTE radio talks to the public internet — nothing touches the customer's IT network.

Can we use our own edge hardware?

Yes. The stack is Docker-Compose. Any 8GB+ x86 box with Linux and a stable network connection runs the same image. We document the canonical hardware so integrators can quote a known-good config; we don't require it.

Standard install BOM

1x OnLogic CL260 hub (8GB / 256GB NVMe)
1x Rittal AX 1045 enclosure (NEMA-4)
1x Mean Well SDR-120-24 PSU (24V/120W)
1x Phoenix QUINT4-CAP supercap (~3 min @ 40W)
1x Teltonika RUT241 LTE router
1x RAK7289V2 LoRaWAN gateway
1x Ruckus T350c WiFi 6E AP
N x WISE-2410 vibration sensors / CT101 clamps

Standard protocols supported

Sparkplug B / MQTT v5
OPC UA
Modbus TCP / RTU
EtherNet/IP
MTConnect
LoRaWAN (CT101, EM300, WISE-2410)
Stage 2 complete — ready for stage 3
3

Week 1-2

First data point

Same shift as install

As soon as the gateway boots and connects to the broker, you see a Sparkplug NBIRTH followed by your first DDATA messages. The OEE snapshot cron starts producing per-shift numbers within 24 hours. The integrator hands the customer a working dashboard before leaving the building — that's the ‘data flowing’ guarantee.

Common questions

What if the data looks wrong?

Often the issue is unit conversion (rpm vs Hz, °C vs °F). The integrator validates each metric against a known-good reading on the line and adjusts in the Functional namespace. No customer engineer time required.

How fast does data update?

Sparkplug Report-by-Exception means data publishes whenever it changes — typically sub-second on running machines. The dashboard polls every 1-5 seconds depending on density preference.

Where does the data live?

Raw events in Convex (Sparkplug B record table). Aggregated metrics in InfluxDB Cloud (90-day retention). Both are tenant-isolated to your enterprise.

Sparkplug B topic stream

spBv1.0/Acme_Foods/NBIRTH/Hub_Massillon
  { bdSeq: 0, seq: 0, timestamp: 2026-05-10T14:23:01Z }
spBv1.0/Acme_Foods/DBIRTH/Hub_Massillon/Filler_Krones_01
  { metrics: [Bottles_Per_Min, Fill_Pressure, State, ...] }
spBv1.0/Acme_Foods/DDATA/Hub_Massillon/Filler_Krones_01
  { Bottles_Per_Min: 412, State: running, seq: 1 }

First OEE snapshot

Filler_Krones_01: OEE 67.3%
A 88.1% × P 79.4% × Q 96.2%
Period: 2026-05-10 06:00–14:00 (1st shift)
Stage 3 complete — ready for stage 4
4

Week 2-3

First action loop

Continuous

Once data is flowing, the alert engine starts evaluating ISA-18.2 alarm rules. The first real alert fires, an AI agent generates a CADET-formatted recommendation, the operator dispatches a work order, and the resolution outcome is captured in the database. That round-trip is the loop — once it closes the first time, the platform compounds from there.

Common questions

How does the AI know which work to recommend?

The recommendation engine joins the alert with asset context (equipment class, last maintenance, FMEA RPN) plus the standards knowledge base (ISO 10816 thresholds, ISA-18.2 priority matrix). It generates the CADET text from a deterministic template — not free-form LLM output.

Can operators override the AI?

Yes. Every recommendation has Acknowledge, Dismiss-with-reason, and Modify. Dismissals feed the learning loop — the system gets less false-positive over time.

What if we already use a CMMS?

Outbound webhooks push work orders to your CMMS (Maximo, Fiix, eMaint, UpKeep) on creation. Status updates can flow back via inbound webhook or polling. Two-way sync keeps both systems consistent.

CADET alert (ISA-18.2 P2)

P2 Filler_Krones_01 — Vibration trend rising
Cause:RMS vibration up 32% over 4-hr trend (4.1 mm/s, ISO 10816 Class II Acceptable→Unsatisfactory)Action:Inspect main drive bearing; review last lube intervalDetail:Z-axis dominant. Sparkplug DDATA from WISE-2410 #4421Effect:Risk of unplanned downtime ≈ $4,200/hr lostTimeline:Respond within 60 min (P2 SLA)

Work order created

WO-2026-0142 — Inspect Filler bearing
  Assigned: J. Marquez (Maintenance, 2nd shift)
  Due: 2026-05-13T22:00Z (60 min)
  Linked: rec_xyz, alert_abc

Outcome captured

Resolved 2026-05-13 21:48
Lube applied; vibration returned to 2.1 mm/s. Saved est. $12,600 of unplanned downtime.
Stage 4 complete — ready for stage 5
5

Week 4+

Customer takeover + ongoing

Continuous

The customer's team operates the platform day-to-day. The integrator handles Tier 2 support (configuration changes, new sensors, custom dashboards). Flowstate handles Tier 3 support (platform issues, major version upgrades). Integrator earns 20% recurring commission for the life of the customer — incentive to keep them successful, not to churn-and-replace.

Common questions

Who owns the data?

The customer. Always. Flowstate stores tenant-isolated data per organization. On termination, customers get a full data export (Convex JSON + InfluxDB CSV) and the system de-provisions cleanly.

What happens at month 12?

Hardware ownership transfers to the customer. Subscription continues monthly with no commitment. Integrator keeps earning commission as long as the customer is active.

How are upgrades handled?

Cloud platform updates are zero-downtime, deployed by Flowstate. Edge stack updates are managed by the integrator on a schedule the customer controls — typically monthly off-shift maintenance windows.

Support tiers

Tier 1 — Customer's own ops team. Day-to-day operations, alert acknowledgement, work-order dispatch.
Tier 2 — Integrator. Adding sensors, building custom dashboards, tuning alert thresholds, ERP/CMMS integration. 24-hr response SLA.
Tier 3 — Flowstate. Platform issues, major upgrades, unusual protocol support, MCP / agent harness config.

Integrator commission

Customer: Acme Foods — 25 machines @ $199/mo
  Monthly recurring: $4,975
  Integrator commission: $995/mo (20%)
  3-yr cumulative: $35,820
Stage 5 complete — ready for stage 6