Action Layer · Shopfloor AI Agents

Issues surfaced at hour 1 — not end of shift.

Production, Quality and IE agents watch every line, every shift. They isolate the exact operator, the exact defect, the exact method correction — and tell your supervisor what to do, before the shift compounds into rework.

Meet the agents
🔍
Quality Agent — Line 3
FPY alert · 07:55 · Hour 1
📉 Low FPY — Line 3. 61% (target 88%). 43 defects of 112 produced.
Where's the issue?
Concentrated, not spread. Op 3: 54% · Op 4: 58%. All others ≥ 87%. Style: HS-442.
💡 Action: 10-min seam-method briefing for Op 3 & 4. Projected FPY recovery: ≥83% by Hour 2.
Live agent · escalation routed
200+
Factories already running Solvei8 on the shopfloor
35K
Machines on platform
7
Countries — South & Southeast Asia and beyond
250M+
Garments manufactured through the platform
The reality on the floor

By end of shift, it's already too late.

Most factories find out about delays, defects and efficiency drops when it's too late to act. The action layer changes that — agents watch in real time and escalate at the first sign of drift.

📉

Defects compound silently

By the time supervisors notice an FPY dip, half a shift's worth of rework is already on the floor. Root cause arrives a day too late.

Efficiency loss is invisible

SAM variance on a single operation can quietly cost $27K a year. Without continuous tracking, no one notices until quarter end.

🕯

Misattribution wastes attention

Supervisors chase the wrong operator because learning curves look like underperformance. The actual bottleneck stays unfixed.

Meet the agents — try it live

Three escalation agents. One operating picture.

Click any agent below to play its live scenario.

🏭
Floor monitoring

Production Agent

Distinguishes learning curves from real underperformance. Surfaces the actual bottleneck.

Issues at hour 1
🔍
FPY monitoring

Quality Agent

Names the operator, identifies the root cause, tells you what to do.

FPY recovered in one shift
Efficiency tracking

IE Agent

SAM vs. actual every hour. Cost of every deviation, down to operator motion.

Catches hidden losses
▶ Live demo — switches when you click an agent
🏭
Production Agent — Line 1
Operator variance · 13:05 · mid-shift
Line 1 throughput · 13:05.
Op 1: 108 pcs/hr ✓ · Op 2: 113 pcs/hr ✓
Op 7: 71 pcs/hr ⚠️ (target 100)
Op 9: 88 pcs/hr ↔ · Op 11: 102 pcs/hr ✓
Op 7 is way off. What's going on?
Op 7 was assigned to waistband attachment 2 days ago — new operation. Learning-curve index: Day 2 of 5-day ramp. Trajectory is normal.
Context: Op 7's previous operation (sleeve join) had a personal best of 118 pcs/hr. Not underperforming — ramping up.
What about Op 9?
Op 9 has a machine tension issue — maintenance ticket #MT-0091 raised 40 min ago, unresolved. That's your bottleneck.
Production · Throughput
Op 7 · Day 2 of ramp
71/hr
Normal — learning curve
Op 9 · machine tension
88/hr
Ticket open · escalation needed
Line 1 average
96/hr
4% below target
The Production Agent separates learning curves from real underperformance — focusing supervisor action on the bottleneck that actually matters, not the one that's most visible.
🔍
Quality Agent — Line 3
First-pass yield drop · 07:55 · Hour 1
📉 Low First-Pass Yield — Line 3. FPY at 07:55: 61%. Target: 88%. Defective: 43 of 112.
Which operations are failing?
Concentrated, not spread. Op 3: FPY 54% · Op 4: FPY 58%. All others ≥ 87%.
Which defects?
Op 3 — Open seam: 9 (64%), wrong stitch density: 5. Op 4 — Open seam: 7 (64%), twisted panel: 4. All style-specific to HS-442.
Are they trained?
Both completed Module 1 only. Module 2 — seam method & stitch spec — scheduled but not done before line start.
💡 Action: 10-min HS-442 seam briefing now. Shadow 30 min. Projected FPY Hour 2: ≥ 83%.
Quality · FPY Alert
Line 3 FPY · Hour 1
61%
↓ 27 pts below target
Op 3 + 4 combined FPY
56%
Open seam + stitch density
Projected Hour 2 FPY
≥83%
After style-method briefing
The Quality Agent breaks FPY down to operator level within minutes of shift start — isolating the exact defect type before it compounds into a costly rework cycle.
IE Agent — Line 2
SAM vs. actual variance · 4-day trend
SAM variance detected. Op 14 (collar attachment), Line 2. Actual: 0.94 SAM vs standard 0.72. Variance: +30%. Duration: 4 consecutive days.
What does that cost us?
At current line output (420 units/day): $76/day excess labour. Monthly: $2,280. Annual if unresolved: $27,360.
What's causing it?
Method analysis: 3 operators using a 2-hand pick-and-place motion. Standard specifies single-hand pre-position (Method 2B). The deviation adds 0.22 SAM/unit.
💡 Recommended: 20-min IE refresher on Method 2B for the 3 operators. Expected recovery to ≤ 0.74 SAM within 2 shifts. Schedule it?
IE · SAM Variance
Op 14 · variance
+30%
0.94 actual vs 0.72 standard
Monthly cost impact
$2,280
4-day trend · 3 operators
Recovery after briefing
≤0.74
Projected within 2 shifts
The IE Agent tracks SAM vs. actual every hour, quantifies the cost of every efficiency deviation, and identifies the exact method correction needed — down to the operator motion.
How it works

Live floor data, in, recommended action out — in seconds.

The Action Layer sits on top of Solvei8's existing MES platform (Tracki8, Maintaini8) and pulls real-time signals from 35,000+ connected machines. The same VDM data spine that powers Command also powers Action.

01 · DATA

Live floor signals

Machine output, operator scans, maintenance tickets, training records — flowing through the MES into the VDM layer in real time.

02 · AGENT

Hourly detection windows

Each agent runs on rolling hourly windows. SAM, FPY and throughput are evaluated against targets, learning curves and historical baselines.

03 · REASONING

Root-cause & recommendation

When a threshold is crossed, the agent generates a structured explanation, identifies the corrective action, and projects the recovery curve.

04 · ESCALATION

Right person, right moment

Supervisors, line leaders and IE engineers receive only the alerts that match their role — and only when action is still possible.

05 · INTERFACE

Chat + dashboard

Same unified interface as the Command Layer. Drill into any alert, ask follow-up questions, or approve recommended actions inline.

06 · FEEDBACK

Closing the loop

Supervisor decisions and outcomes feed back into the agent — improving recommendation accuracy and reducing false escalations over time.

Walk through a live shift, for your factory.

We'll connect a sandbox to your line data and show what hour-1 issue detection looks like on your actual production.