Operations & Supply · App 18
Every signal from the field becomes an accountable action
A planned app for operations-heavy and sovereign organisations: frontline work in Indian languages, offline sync, IoT and camera signals routed into maintenance, safety and store tasks, with budgeted GPU compute.
One app that connects the physical side of a business (places, workers, devices, cameras, screens, on-prem boxes, GPUs and later robots) to the governed Action layer of the One3 business OS. Every signal from the field becomes an accountable business action with an owner, an approval where needed, an audit trail and a measured outcome. It is designed to work offline, in Indian languages and, when the buyer requires it, fully inside the buyer's own premises.
The physical side of the business OS
Operations teams usually run the physical world on separate products: an IoT platform that stops at dashboards, a video analytics box that sends alerts to a WhatsApp group, a checklist app priced per seat, a maps API wired into a CRM by an integrator, a signage player with its own console, and GPU capacity rented on whoever's card is open. Each product detects something, and then a person retypes it into the system that does the work.
Edge, IoT & Compute is planned as one app with ten modules that closes that gap. An IoT rule, a vision detection, a failed checklist item or a geofence exit calls typed Actions in other One3 apps: a work order in Assets & Maintenance, an observation in EHS, a request in Facilities, a ticket in Service Desk, a task for the site team. The Action Bus applies risk tiers, reversibility classes and approvals, and every step is audited.
How it fits with the rest of One3
- Edge Sync is the offline substrate. Frontline, field service, POS and AR work offline through declared sync scopes, and queued Actions are re-checked for permission and policy when they replay.
- Maps & Location gives every app one location model: territories, beats, routes and visit verification on the same party and address records CRM and Field Service use.
- Frontline Worker App puts tasks, checklists and notices in the worker's language, by voice, photo or WhatsApp, without forcing an app install.
- IoT Hub and Vision AI turn telemetry and existing cameras into reviewed proposals, not dashboards nobody watches. Vision AI runs at the edge, sends metadata only and keeps face identification off.
- Kiosk & Digital Signage drives screens from live prices, KPIs and emergency events.
- One3 Box Manager runs sovereign and edge appliances with signed, customer-approved updates.
- One3 Compute gives teams budgeted GPU jobs with hard caps, and posts every job's cost to AI FinOps.
- Spatial Work Instructions (AR) and Robotics Bridge are later-wave modules for guided procedures and robot missions with safety kept in the vendor's system.
Guardrails built in
No automated adverse decision about a person: detections, geofence exits and missed checklists are signals for a human. Location tracking runs only on shift, after notice. No physical equipment control before the physical-risk tier planned for W6. Customers control their own boxes, and a licence expiry degrades to read-only, never to data lockout. Outcomes such as downtime avoided, checklist compliance and travel time saved are designed to be written to the Investment & ROI Ledger against baselines frozen at onboarding.
The app is planned in waves: W4 (Year 3) for Maps, Edge Sync and Frontline; W5 (Years 4–5) for IoT Hub, Vision AI, Signage, One3 Box Manager and One3 Compute; W6 (Years 6–10) for AR and Robotics.
Problems solved
What Edge, IoT & Compute fixes
Checklists on paper or WhatsApp groups
Store and site checks are skipped and nobody knows until an audit fails. Planned: scheduled checklists with photo evidence, AI photo verification, voice answers in Indian languages and failed items that become corrective tasks.
Contract workers without seats or apps
Casual and contract staff cannot be given seats or asked to install an app. Planned: a WhatsApp mode with no install, shared-device mode and frontline pricing options per active site or worker-month.
Work lost where there is no signal
Technicians in basements and villages lose work in online-only apps. Planned: declared sync scopes, an offline Action queue with idempotency keys, conflict rules, a review queue and site edge nodes.
Territories and beats in spreadsheets
Reps overlap and travel too much. Planned: territories, beats, route optimisation and visit verification on the same records CRM and Field Service use.
Sensor alarms nobody acts on
IoT platforms stop at alerts and the machine still breaks. Planned: no-code rules that create work orders, frontline tasks or EHS observations through Actions, with acknowledgement and escalation.
CCTV used only after an incident
Hundreds of cameras, no workflow. Planned: camera-agnostic edge inference on existing RTSP/ONVIF cameras, with detections reviewed and turned into tasks, face identification off by default.
Screens showing last month's offer
Signage is managed apart from pricing and POS. Planned: playlists with mandatory content expiry, POS-synced menu boards and site-wide emergency override.
Data must stay on premises, and GPU spend is invisible
Regulated buyers need AI without data leaving the building, and fine-tunes run on untracked accounts. Planned: One3 Box with local models and signed updates, and quota-bound GPU jobs with cost posted to AI FinOps.
Key capabilities
What you can do
Signal to governed action
IoT rules, vision detections, geofence events and failed checklist items call owner-app Actions with risk tiers, approvals and audit.
Works offline
Declared sync scopes, queued Actions re-validated on replay, conflict rules that never discard the losing value, and site edge nodes for WAN outages.
Voice-first, WhatsApp-first frontline
Checklists, tasks and notices by voice in Indian languages, by photo, or over WhatsApp with no app install.
Privacy-first Vision AI
Edge inference on existing cameras, privacy masks, metadata-only upstream, purge schedules and face identification off by default.
One location model
Territories, beats, routes, geofences and visit verification on the same party and address records used by CRM and Field Service.
Sovereign when required
One3 Box runs the platform on premises or air-gapped, with signed bundles, customer-approved updates and consented diagnostics.
Budgeted GPU compute
Cost estimates before submit, hard quotas that cannot become soft caps, residency-aware pools and JobCost posted to AI FinOps.
Screens from live data
Playlists with mandatory expiry, POS-synced menu boards, live KPIs and site-wide emergency override.
Outcomes on the ROI Ledger
Downtime avoided, checklist compliance, detections closed and travel time saved are designed to be measured against baselines frozen at onboarding.
Modules
10 modules, one data model
Core modules come with the app. Add-ons switch on when you need them. Some modules can be bought on their own.
Maps & Location
One location model for territories, beats, routes and verified visits
ExploreEdge Sync
Offline work that never loses or duplicates a business action
ExploreFrontline Worker App
Tasks, checklists and notices by voice, photo or WhatsApp, in the worker's language
ExploreIoT Hub
Sensors and machines that create work orders, not unwatched alarms
ExploreVision AI
Existing cameras as privacy-preserving site signals that drive work
ExploreKiosk & Digital Signage
Screens and kiosks driven by live prices, KPIs and emergency events
ExploreOne3 Box Manager
Run One3 on premises or air-gapped, with updates you approve
ExploreOne3 Compute
Budgeted, auditable GPU jobs with hard caps and cost per job
ExploreSpatial Work Instructions (AR)
Capture an expert's repair once and guide every technician step by step
ExploreRobotics Bridge
Robot missions from business events, with safety kept in the vendor's system
ExploreWorkflows
How the work flows
End-to-end workflows inside the app and across One3.
Scheduled checklist to corrective action
A schedule opens a checklist run, the worker answers by voice, tap or photo, and failed items become owned corrective tasks with escalation.
- 1.A schedule generates a checklist run for a site, shift, asset or zone
- 2.The worker opens it in the app, on WhatsApp or on a shared device
- 3.Items are answered by voice, tap or photo; photo items are checked by AI against the requirement
- 4.Low-confidence photo results go to a human review queue
- 5.Failed or out-of-range items create a corrective task with owner and due time
- + 2 more steps →
Telemetry to business action
Device telemetry is decoded and evaluated against no-code rules; a sustained condition raises one deduplicated alert and proposes a work order, task or EHS observation.
- 1.A device or gateway sends telemetry over MQTT or HTTP with its own identity
- 2.The decoder normalises payloads and units; telemetry is stored with rollups
- 3.Rules evaluate thresholds, rate of change, silence and duration
- 4.A met condition raises a deduplicated IoTAlert with severity
- 5.If the rule has an Action, the Action Bus applies risk tier, DoA and approvals
- + 2 more steps →
Camera onboarding with privacy gate
No camera runs inference until a privacy profile is approved, and no detection drives an Action until the site validation gate passes.
- 1.Discover or add the camera on the site edge node and map it to site and zone
- 2.Complete the privacy profile: purpose, lawful basis, masks, retention, allowed use cases
- 3.High-risk use cases go through a DPIA in Privacy Ops
- 4.A privacy approver reviews; notice signage is recorded
- 5.Inference starts in review-required mode
- + 2 more steps →
Offline capture and replay
Work done offline is queued as Actions and replayed on reconnect with permission, policy and DoA checked again.
- 1.The user acts offline; the app validates against a cached policy snapshot
- 2.The write is queued as an Action with idempotency key, clock and dependencies
- 3.On reconnect the queue uploads in dependency order
- 4.The Action Bus re-checks entitlement, permission, policy and DoA
- 5.Rejected Actions return with a reason and are kept for review
- + 2 more steps →
Signed update to a sovereign box
One3 publishes a signed bundle; the customer verifies, approves and installs it with pre-checks, health gates and automatic rollback.
- 1.One3 publishes a signed bundle with SBOM on a release channel
- 2.Connected boxes receive it on the channel; air-gapped sites import through the transfer gateway with scan, checksum and dual approval
- 3.Signature and SBOM are verified on the box
- 4.The customer reviews release notes and approves
- 5.Pre-checks confirm backup age, capacity and health
- + 2 more steps →
GPU job with quota and residency
A job is estimated before it starts, blocked if it could exceed its quota, placed only in residency-compatible pools and costed when it ends.
- 1.Submit from a template in AI Lab, Video Factory or the API
- 2.Dataset access and consent checks run
- 3.Cost and GPU-time are estimated with an upper bound
- 4.If the upper estimate exceeds remaining quota, the job is blocked and a quota raise goes to DoA approval
- 5.Residency and classification policy selects an eligible pool
- + 2 more steps →
AI inside
AI skills that act through governed Actions
Every AI skill runs through the same typed Actions people use, with risk tiers, approvals and an audit trail.
Photo verification A3
Checks a checklist photo against the item requirement with a confidence and short reason. Below-threshold results always go to a human; safety-critical items stay human.
Voice-first checklist interaction
Workers speak to open tasks, answer items and add notes in Indian languages, with a read-back before submit. Human-invoked.
Route optimisation with explanation A3
Multi-stop routes with time windows, skills and capacity, listing unassignable jobs and the binding constraint in plain words. Assignment stays with the dispatcher.
Rule drafting from plain language A2
Turns 'tell me when the chiller stays above 8 °C for 20 minutes' into a condition, window, suppression and target Action, simulated on history. Enabling stays human.
Telemetry anomaly detection A3
Seasonal per-asset baselines, change points, drift and stuck-sensor patterns, raised as proposals.
Edge vision inference A3
PPE, restricted-zone presence, queue length, shelf gap and similar use cases run on the edge node; actions only after the site validation gate.
Signage content drafting and pre-check A2
Drafts content within brand rules and checks expiry, POS price binding and data classification before publish. Customer-facing publish needs approval.
Release impact summary A1
Summarises CVE delta, migrations, breaking changes and rollback target for the customer approver of a box update. Summary only, no approval authority.
Compute cost estimate and failure triage A3
Estimates job cost with an upper bound before submit and classifies failures (out of memory, data error, preemption, quota stop) with the smallest fix.
Expert capture to procedure draft A2
Turns a recorded repair into ordered steps with tools, parts and warnings; the expert edits every step before approval.
Digital Employees
Agents that operate Edge, IoT & Compute
Rent them by the month. They work under your team's supervision and their cost and returns appear in the Investment & ROI Ledger.
AI Store Operations Assistant
Role #131. Shelf-gap reports, store task lists and shift handover notes from Frontline and Vision AI, under the store manager's supervision.
Learn moreAI Field Service Dispatcher
Role #68. Scheduling, routing and parts checks using Maps & Location route proposals; dispatchers approve assignments.
Learn moreAI Maintenance Planner
Role #67. Drafts maintenance schedules and work-order proposals from meter and IoT Hub signals for the maintenance planner.
Learn moreAI Dispatch & Trip Planner
Role #137. Load planning, route optimisation and driver assignment proposals for logistics teams.
Learn moreAI Operations Manager
Role #61. Exception roll-ups across sites, task allocation and SOP adherence checks.
Learn moreAI Site Safety Monitor (proposed)
Proposed Role Factory template. Prepares the vision detection review queue, drafts EHS observations and weekly PPE and zone summaries; never targets individuals.
AI Shift Supervisor Assistant (proposed)
Proposed template. Chases overdue items, compiles handovers and reports acknowledgement gaps for supervisors.
AI IoT Reliability Analyst (proposed)
Proposed template. Tunes noisy rules, investigates silent devices and drafts root-cause notes; enabling rules stays human.
Integrations
Connects with the tools you already use
- WhatsApp Business Platform
- SMS gateway
- Voice channel
- OpenStreetMap vector tiles
- MapmyIndia Mappls
- Google Maps Platform
- Telematics and fleet-tracking providers
- DigiLocker
- UPI via Payments Hub
- AWS IoT Core and Greengrass
- Azure IoT Hub, IoT Central and IoT Operations
- ThingsBoard
- Partner OT gateways (Modbus, BACnet, OPC UA)
- LoRaWAN network servers
- ONVIF and RTSP cameras
- VMS and NVR systems
- GPU cloud providers
- Active Directory, LDAP and SAML
- Customer HSM and KMS
- ARKit, ARCore and WebXR
- Robot fleet managers and ROS 2
Connector availability depends on provider access and is verified before launch. How integrations work →
Security & compliance
Built to be trusted
No automated adverse decisions about people
Vision detections, geofence exits, missed checklists and visit results cannot drive disciplinary, pay, rating or entry decisions. There is no Action that does it and no field that stores one.
No face identification
No face templates, embeddings or biometric templates are stored. Any later evaluation would need a DPIA, counsel sign-off and an R4 admin Action, and is never offered for law-enforcement or public surveillance.
Tracking only on shift, with notice
Location sampling starts only during an active shift for roles whose policy enables it, stops at shift end even offline, and traces are purged on schedule. Purge keeps running after cancellation.
Least authority for things
Every device, camera, screen and box has its own credential with per-device topic ACLs, rotation and revocation. Things are never granted workforce permissions.
Customer control of sovereign boxes
Updates, diagnostics exports and support sessions need customer approval; support sessions are time-boxed and recorded; licence expiry degrades to read-only and export stays available.
Regulatory pointers
DPDP Act duties for video, voice and location, CERT-In readiness and drone rules are mapped to controls as pointers for counsel. Verify with counsel; this is not legal advice.
Industry Editions
Industries that use Edge, IoT & Compute
Frequently asked questions
Is Edge, IoT & Compute available now?
No. It is outside the Year-1 boundary. Maps & Location, Edge Sync and the Frontline Worker App are planned for W4 (Year 3); IoT Hub, Vision AI, Kiosk & Digital Signage, One3 Box Manager and One3 Compute for W5 (Years 4–5); AR and Robotics Bridge for W6.
Do I need to buy this app to use POS or Field Service offline?
The plan's proposed decision is no. The sync runtime that first-party offline modules need is included with those modules. The Edge Sync admin console, tenant-defined scopes and site edge nodes come with this app.
Are devices, cameras and screens counted as seats?
No. Things are never seats. Candidate units under evaluation include per site with included allowances, per camera stream, per screen or per device band. No prices are set in the plan.
Does Vision AI recognise faces?
No. Face identification is off by default and not offered in W5. Inference runs at the edge, only metadata and blurred evidence go upstream, and detections describe site conditions rather than identified individuals.
Can IoT Hub control machines?
Not before the physical-risk tier planned for W6. Until then, commands are limited to non-safety device configuration and diagnostics. Robot missions always go through the vendor's fleet manager, whose safety system stays authoritative.
Do Digital Employees replace our supervisors or technicians?
No. Digital Employees work for your team under supervision. They draft routes, tasks, rule proposals and summaries; people approve publishing, safety-critical items, box updates, quota raises and robot missions.
Can One3 run inside our own data centre?
That is the plan for One3 Box (W5): the platform on premises or air-gapped with local open-weight models, signed updates you approve, and diagnostics that leave only after you preview and approve them. It requires the Enterprise Sovereign isolation tier.
Works with
Related apps
Platform
One governed foundation for every human, AI agent, integration and channel
ExploreCRM & Sales
A WhatsApp-first CRM where sellers and Digital Employees work the same pipeline
ExploreOperations
Buy, stock, make, move, maintain and service physical goods on one Item master
ExploreCommerce
One catalogue, one stock position and one order book for every place you sell
ExploreLegal, Risk & Compliance
Contracts, consent, controls and risk as governed work on the data you already run
ExploreFacilities & ESG
Buildings, visitors, safety, energy, carbon and leases on one traceable data model
ExploreIT & Security
IT service, devices, security hygiene and engineering visibility on your people data
ExploreIntelligence
Cited answers, one set of metrics and decisions tracked to a measured result
ExploreAI Workforce
Rent ready-made AI Digital Employees that work for your team, with a monthly ROI statement
ExplorePlan your Edge, IoT & Compute rollout with us
Tell us about your team and the outcomes you want. We will walk you through the apps, the Digital Employees and the plan for your industry.