Manufacturing IoT Connectivity Management with Kajeet SentinelOS

The 2 a.m. Call That Tells You Something Is Wrong with Your Tooling

Here is a story that will sound familiar to anyone running a connected plant. A vibration sensor on a stamping press goes quiet at 11 p.m. Nobody notices, because noticing requires a person to be looking at a dashboard, and the person who looks at that dashboard went home at six. The predictive maintenance model that depends on the sensor quietly starts working from stale data. At 2 a.m. the press develops the exact bearing fault the sensor existed to catch. Third shift calls the maintenance lead, the maintenance lead calls IT, and IT spends forty minutes logging into a carrier portal to learn what a machine could have reported instantly: the sensor’s cellular gateway dropped off the network three hours ago.

The downtime was never really a connectivity failure. It was a workflow failure. The information existed the whole time, locked behind a login screen that no one was watching.

Manufacturers have spent a decade instrumenting everything: presses, chillers, forklifts, trailers, tanks. The connectivity that carries all that telemetry is still mostly managed the way it was in 2015, through portals built for humans with mice. Kajeet built SentinelOS to close that gap. It makes cellular connectivity programmable, and just as important, it makes connectivity something an AI agent can watch, reason about, and act on around the clock.

What SentinelOS Actually Is

SentinelOS is the agent-ready operating layer on top of Kajeet’s Sentinel IoT platform, the same platform Kajeet customers already use to manage multi-carrier IoT deployments. Launched in June 2026, it opens Sentinel up through four surfaces, each aimed at a different kind of user:

REST API: 179 operations across 200 OpenAPI paths. This is the foundation for ERP, MES, and ITSM integrations.
Sentinel MCP server: 11 grouped tools covering 69 actions, currently in public beta. Any MCP-compatible assistant — Claude, ChatGPT, Codex, Gemini, Cursor — can query devices, analyze usage, review invoices, and manage failover with natural language.
Sentinel CLI: a one-line curl install, or npm if you prefer. Built for IT admins who would rather pipe device status through jq and cron than open a browser.
Agent Skills: open playbooks that teach AI coding tools like Claude Code and Codex how to run multi-step IoT operations correctly.

The design principle behind all four: the console is where you look things up, not where the work happens. For a plant with thousands of connected endpoints, the work should happen in code and in agents.

What This Looks Like on a Factory Floor

1. An Agent That Watches Your Sensors So Nobody Has To

Use case · IIoT sensor and gateway triage

Connect an AI agent to the Sentinel MCP server and the 2 a.m. story changes shape. The agent checks device health on a schedule, compares current status against yesterday’s, and knows the difference between a gateway that reboots nightly by design and one that has actually gone dark. When the stamping press gateway drops off at 11 p.m., the agent notices within minutes, pulls the device details, checks whether the whole site lost signal or just one unit, and posts a summary to the maintenance channel with the device name, last-seen time, and its best guess at the cause. If a support case is warranted, it can open one in ServiceNow through the same MCP connection, with the diagnostic context already attached.

Here is roughly what that interaction looks like from the operator’s side:

Which gateways at the Dayton plant went offline in the last 12 hours?

Two devices dropped off the network at DAYTON-PLANT-02:

1. GW-PRESS-LINE3 — last seen 11:04 p.m.

2. GW-CHILLER-EAST — last seen 2:47 a.m.

GW-PRESS-LINE3 shows normal signal history before the drop, so this looks like a device-side issue rather than coverage. Want me to check its usage pattern for the past week or open a support case with these details?

Used tools: sentinel_devices · sentinel_usage

2. Failover That Manages Itself Between Failures

Use case · Production line and site failover

Most plants treat cellular as the backup path for wired broadband, then never test it until the day it matters. Kajeet’s SmartFailover handles the switchover itself, and the SentinelOS smart_failover tool exposes ten read-only actions around it: device telemetry, live connection status, signal history, usage, readiness checks, failover event history, and fleet-wide summaries across the customer hierarchy. That turns failover from a thing you hope works into a thing an agent verifies weekly. A scheduled task can confirm every failover unit at every site is healthy and passing its readiness check, then report the exceptions. When a primary circuit does go down, the same agent can pull the event log and tell you exactly when the switch happened and what it looked like from the network side.

3. The Pre-Shift Sweep: A Script That Checks Every Gateway Before First Shift Clocks In

Use case · Scripted device checks with the Sentinel CLI

Not every plant wants an AI agent, and not every job needs one. Plenty of manufacturing IT teams live in scripts, and for them SentinelOS ships a proper command line tool, installable with a one-line curl script or npm. The pattern that pays off first is the pre-shift sweep. A cron job runs at 5 a.m., pulls device status across every site with the sentinel CLI, pipes the output through jq to keep only the exceptions, and posts the result to the plant IT channel in Teams. When first shift walks in at six, the team already knows whether every gateway on every line is answering, and the one that is not has a name and a last-seen time next to it. Nobody logged into anything.

The same approach catches quieter problems. A weekly script that ranks devices by data consumption will surface the PLC gateway that started chattering after a bad firmware update long before it drains the data pool, and a monthly one can diff the active device inventory against the asset register so connectivity records stop drifting away from reality. Because the CLI is just another command, these checks slot into the cron jobs, CI pipelines, and monitoring stacks a plant already runs. The health of the whole IoT deployment becomes one more thing your existing automation watches, not another browser tab.

4. SIMs That Activate Themselves When the ERP Says Ship It

Use case · OEM machinery provisioning

Equipment manufacturers who ship machines with embedded cellular have historically needed a human to activate each SIM after the sale, in a portal, from a spreadsheet. With the Sentinel activation API, provisioning becomes a step in the order workflow: when the ERP marks a machine Delivered, the integration activates the line. The activation tools support async operation with a preview and confirm step, and Kajeet provides test IMEIs so you can dry-run the whole flow before a single real machine ships. Suspension works the same way in reverse, so a machine that comes back for refurbishment stops billing the day it arrives.

The Part Your Security Team Will Ask About

Every conversation about agents touching operational systems arrives at the same question: what stops the AI from doing something destructive? SentinelOS has a specific, checkable answer rather than a reassuring paragraph.

Agents authenticate the same two ways humans and systems do. Interactive AI clients use OAuth 2.1 with PKCE, discovered automatically, so a plant engineer connecting Claude to Sentinel signs in with their own credentials and the agent inherits exactly their permissions, nothing more. Headless automations use scoped API keys. Either way, an agent working through Sentinel MCP sees only the device groups its user is authorized to see.

Then there is the confirm step. Any destructive action — suspending a line, activating a SIM, swapping a SmartSIM profile — does not execute on first request. The tool returns a preview of what is about to happen, and the change only runs when it is explicitly confirmed. Every action, read or write, lands in the audit log. Your agent can be curious all day long; it cannot be careless.

Connecting an Agent Takes About Five Minutes

This is the entire configuration for Claude Desktop. Other MCP clients are equally short:

claude_desktop_config.json

{
  "mcpServers": {
    "sentinel": {
      "url": "https://mcp.kajeet.dev/mcp",
      "transport": "streamable-http"
    }
  }
}

Add the server, sign in through the OAuth window with your normal Sentinel credentials, and start asking questions. For CI pipelines and cron jobs, generate an API key in the developer dashboard and call the same endpoint with an X-API-Key header. The full walkthrough lives in the quickstart guide, and there is a step-by-step Field Note for Claude.ai that requires no code at all.

SentinelOS vs. Legacy Carrier Portals

Capability	Legacy carrier portals	Kajeet SentinelOS
Device management method	Manual dashboard logins	REST API, CLI, MCP server, Agent Skills
AI agent compatibility	Not supported	Native via Sentinel MCP: 11 tools, 69 actions
ERP and ITSM integration	Custom workarounds, if any	179 operations across 200 OpenAPI paths
Multi-carrier visibility	One portal per carrier	Single control plane across networks
Failover management	Separate hardware tooling	SmartFailover telemetry, events, and readiness via API and MCP
Troubleshooting model	Reactive, human-triggered	Continuous agent monitoring with human-confirmed actions
Governance and audit	Limited, manual review	OAuth 2.1 + PKCE, scoped keys, preview/confirm, full audit log

How to Roll This Out

The teams that get value fastest do not start with a grand automation program. They start with one annoying chore.

Map your current touchpoints. List every portal, login, spreadsheet, and weekly report your plant IT and logistics teams touch to manage connectivity. This list is your automation backlog.
Pick your first surface. If you have developers, the REST API and your ERP is the highest-leverage integration. If you have IT admins, start with the CLI. If you have neither to spare, connect an AI assistant through MCP, because it needs no code at all.
Set up governed access. Sign agents in through OAuth so they inherit existing user permissions, or issue scoped API keys for headless jobs. Decide up front which roles are allowed to confirm destructive actions.
Automate the noisiest chore first. For most plants that is the morning offline-device sweep or the monthly usage anomaly hunt. One scheduled agent task, visible results in week one.
Then scale. ERP-driven provisioning, weekly SmartFailover readiness checks across sites, tracker silence detection, automated support cases. Each one retires a manual workflow.

The Numbers That Matter

179OpenAPI REST operations for ERP and ITSM integration

11 / 69MCP tools and actions available to AI agents

<5 minfrom MCP config to first successful agent call

10SmartFailover actions, from signal telemetry to fleet readiness

Source: Kajeet Developer Platform documentation at kajeet.dev, July 2026. Sentinel MCP and CLI are in public beta.

Frequently Asked Questions

What is Kajeet SentinelOS and how does it apply to manufacturing?

SentinelOS is the agent-ready operating layer for connected devices, built on Kajeet’s Sentinel IoT platform. It gives manufacturers four ways to manage cellular IoT devices without a dashboard: a REST API with 179 documented operations, the Sentinel MCP server for AI agents, a CLI for scripting, and Agent Skills for AI coding tools. Sensors, asset trackers, and connected machinery get managed as code and watched by agents instead of humans babysitting portals.

How does SentinelOS reduce downtime for smart factory sensors?

An agent connected through Sentinel MCP checks device health continuously, flags offline gateways and usage anomalies, pulls SmartFailover telemetry, and can open support cases with diagnostics attached. Problems get caught in minutes instead of waiting for someone to look at a dashboard, which usually means they get fixed before the production line notices.

Can SentinelOS integrate with our existing ERP system?

Yes. The REST API exposes 179 operations across 200 OpenAPI paths covering activation, suspension, usage, and inventory. A common pattern is wiring SIM activation to the order workflow so a machine’s connectivity goes live the moment the ERP marks it shipped. Kajeet’s docs target a first successful API call in under five minutes, and test IMEIs let you rehearse activation flows safely.

Is it safe to let AI agents manage manufacturing connectivity?

Sentinel MCP runs under Kajeet’s existing identity, policy, and audit controls. Agents sign in with OAuth 2.1 with PKCE or scoped API keys and inherit only the permissions of their user. Destructive actions return a preview first and require explicit confirmation before executing, and every action is logged. The agent operates with your badge, under your rules, on the record.

Does SentinelOS support multi-carrier connectivity for supply chain tracking?

Yes. Sentinel manages devices across multiple carrier networks from a single control plane, and SentinelOS makes that programmable. Agents and scripts can compare device performance and usage across networks and watch SmartFailover status for assets and sites in different coverage zones.

Do we need developers to get started?

No. The MCP path requires no code: add the server URL to Claude, ChatGPT, or another MCP-compatible assistant, sign in with your Sentinel credentials, and start asking questions about your IoT deployment. Developers become valuable when you want deeper integrations, like ERP-driven provisioning or custom agent workflows.

Where This Is Heading

The interesting shift is not that connectivity management got an API. It is that the day-to-day operation of an IoT deployment is becoming something you delegate rather than something you do. The dashboard era asked a person to notice problems. The agent era asks a person to approve fixes. For a manufacturer, that difference shows up in concrete places: maintenance windows that do not slip because a sensor was silently dead, failover gear that gets verified weekly instead of discovered broken, provisioning that keeps pace with shipping instead of trailing it by days.

Kajeet has run managed multi-carrier connectivity for education, healthcare, enterprise, and public sector deployments for years. SentinelOS is that same platform with the door opened to code and to agents, under the same governance customers already trust. The public beta is live now at kajeet.dev, and a free trial SIM kit gets you from reading about this to testing it with real hardware.

Put an Agent on the Night Shift

Order a free trial SIM kit, connect your AI assistant to Sentinel MCP, and run your first device query before your next production meeting. The quickstart is the five-minute version.