OFFLINE ATTENDANCE FOR CONSTRUCTION CREWS/Evidence before payroll, not certainty invented afterward.
Construction attendance
that keeps working offline.
GroundNerve records attendance on workers’ Android phones, even without internet. It preserves the underlying evidence and flags uncertain hours before payroll.
11 workers · 200 worker-days · Jun 23–Jul 23 · 10 signed
Live feed
LIVE FIELD TRUTHSIGNAL CAPTUREDPROOF VERIFIEDOWNER MISSINGHANDOFF AT RISKAPPROVAL STUCKPAYROLL FOGACTION READYHUMAN GATELOOP CLOSEDLIVE FIELD TRUTHSIGNAL CAPTUREDPROOF VERIFIEDOWNER MISSINGHANDOFF AT RISKAPPROVAL STUCKPAYROLL FOGACTION READYHUMAN GATELOOP CLOSED
Inside the screen
AI knows what the company typed.
EmailTicketsCalendarsReportsDashboards
Outside the screen
It does not know what happened.
Who's there?
What changed?
What's stuck?
Proof?
Approval?
Next move?
The physical world has no clean API. That is the missing layer.
The proof, not the pitch
One disputed shift. Every source exposed. No black box.
SENSE → VERIFY → GATE → RECEIPTFLORIN I. · VILA SNAGOV · TODAY
>
Observed facts, derived confidence, AI recommendation, human decision.Money and risk never cross the gate alone.
The missing infrastructure
Not another brain. A nervous system.
GroundNerve is the evidence and action layer between field operations and any AI agent. It does not ask the model to guess reality. It makes reality inspectable.
- 01NervesField signals arrive with scars.raw + local-first
Phone ledger, GPS, device health, offline sync, supervisor confirmation.
- 02TruthEvery claim earns provenance.observed + derived
Source, timestamp, freshness, confidence, permission, uncertainty.
- 03AgentThe model gets tools, not vibes.recommend + prepare
Explain the exception, cite evidence, prepare the smallest safe next move.
- 04BrakesRisk stops for a human.approve + reject
Payroll, money, people, customer, legal, material, and safety impact.
- 05ReceiptThe loop closes with evidence.audit + follow-up
Who decided, what they saw, what changed, and what remains unresolved.
Tool surface · MCP
Agents do not scrape screens here. They call tools — and every tool carries a permission tier the model cannot cross alone.
get_ops_snapshotToday’s field state, sourced.readexplain_attendance_dayOne worker’s day, evidence cited.readdraft_payroll_correctionPrepared. Never sent alone.draftexport_payroll_csvMoney. Stops for a human.human gate
FIRST WEDGEPayroll-safe field truth
One worker. One site. One day. One decision the owner can defend.
Everyone is trying to make AI smarter.
Bigger models. More context. More tools. The same missing reality.
We made sand think. Then locked it behind a blinking cursor.
Reality moves first. The record comes later.
Managers reconstruct the day from fragments.
A phone call here. A photo there. A spreadsheet edited after payroll.
The physical world does not expose a clean API.
The bottleneck is not intelligence.
It is|
GroundNerve gives agents a nervous system for field operations.
|
Agentic action
Visibility is not management. The agent runs the loop.
A useful agent does not stop at an answer. It finds the exception, cites the evidence, prepares the move, stops at risk, and follows up until reality changes. AI does the chasing. Humans keep control.
01
Sense: the AI watches shifts, routes, proof, approvals, and risk.
02
Decide: it finds the exception, explains the cause, and picks the next move.
03
Gate: money, payroll, people, and client risk stop for human approval.
04
Act: it uses tools, routes owners, sends requests, and keeps chasing until closed.
One case, full loop
Florin's uncertain shift becomes a defensible decision.
- Sense
Local check-in lands before the network returns.
- Verify
GPS becomes stale. Device health explains the silence.
- Gate
The agent asks the supervisor and keeps 1h 42m outside payroll.
- Close
The owner approves safe hours. The receipt preserves every source and choice.
You decide what the agent can do.
← default: approval before riskProof Room
If the evidence cannot survive a challenge, the action does not deserve to happen.
This is the first wedge: payroll-safe field truth. The broad vision earns the right to exist by proving one hard decision first.
DECISION RECEIPT · GN-4417DESIGN-PARTNER INTERFACE
Q · Which hours are defensible enough to enter payroll today?
6h 18m are safe. 1h 42m stays out.
SourceOfflineV2 ledger
Freshness4m ago
GPS confidencemedium
Devicebackground limited
Supervisorpending
Permissionhuman-gated
The recommendation cannot approve itself. The supervisor ask is prepared.
Human-gated- 01Observed
phone ledger + site GPS
- 02Derived
confidence + uncertainty
- 03Recommended
approve safe, hold uncertain
- 04Human approved
owner · 18:06
- 05Executed
receipt + follow-up
The founder contract
AI should be powerful enough to act. And unable to lie about why.
We refuse both extremes: a passive dashboard that only reports damage, and a black-box agent that moves money without proof.
Evidence, not surveillance
Every claim carries source, freshness, confidence, and permission.
No blind autonomy
AI can watch, explain, prepare, and ask. Money and risk stop at a human gate.
Receipts, not vibes
Every sensitive action records what the agent saw, what the human approved, and what changed.
CONSTRUCTION IS THE WEDGE, NOT THE LIMIT
If work happens outside the laptop, AI needs a way in.
ConstructionCleaningSecurityMaintenanceUtilitiesTelecomInstallationLogistics
Design partner cohort
This category will not be built from a deck.
It will be built with operators who carry the scar tissue and AI builders who refuse to ship another blind agent.
Bring us one ugly, expensive field decision. We will make the evidence visible, the action safe, and the outcome defensible.