English 
Français 
Español 
Deutsch 
Português 
Italiano 
العربية OBSERVE
The physical world
- Sensors and meters
- Machines and actuators
- Buildings and fleets
- Video — coming soon
Physical AI infrastructure
The operating system for AI in the real world
Language models can reason. Kilo gives them a governed way to observe live infrastructure, operate connected devices, and verify what happened — across one sensor or thousands of sites.
Built-in AI assistant · Open MCP server · Cloud or on-premise
One operational layer
From physical signal to safe, verified action
Kilo normalizes fragmented hardware into one permissioned system that an AI can understand and operate without bypassing production controls.
GOVERN
Kilo execution layer
- Digital device models
- Rules and commands
- Permissions and approvals
- Audit and verification
REASON
AI and software
- Kilo AI Assistant
- MCP-capable agents
- REST and gRPC services
- Real-time consumers
Use the AI you choose
One platform, two ways for AI to work
Use the assistant built into Kilo or bring the AI client your team already trusts. Both operate through the same platform permissions and organization boundary.
INSIDE KILO
The built-in AI Assistant
Ask about live and historical telemetry, provision devices, build and simulate rules, and configure alarms in plain language.
- ✓Grounded in your live deployment
- ✓Confirms consequential changes
- ✓Bring OpenAI, Anthropic, Ollama, or a compatible model
YOUR AI CLIENT
Any MCP-capable agent
Connect ChatGPT, Claude, Codex, Cursor, or another MCP client. Sign in with your Kilo account; the agent inherits exactly your permissions.
- ✓OAuth sign-in — no token copying
- ✓Organization-scoped access
- ✓Open Streamable HTTP endpoint
Production controls
AI action without giving up operational control
Physical actions need more than a clever model. Kilo supplies the controls between a request and a real device.
01
Permission inheritance
An agent can never see or change more than the signed-in user is allowed to.
02
Human confirmation
Consequential and destructive changes pause for explicit approval.
03
Test before deploy
Rules are validated and simulated before they touch live infrastructure.
04
Version and rollback
Every automation change is versioned and a known-good build is one click away.
05
Typed, verified commands
Parameter bounds constrain actions; optional verification confirms the device responded.
06
Forensic history
Commands, deployments, permission changes, and outcomes remain auditable.
Built for operations
The same execution layer across every physical domain
CO₂ · LEAKS · ENERGY · ACCESS
Facilities and buildings
Understand conditions across sites, coordinate alarms, and command connected HVAC, valves, gates, and controls.
VIBRATION · POWER · UPTIME
Factories and infrastructure
Reason across vibration, temperature, power, and machine state before a rule or operator takes action.
GPS · CAN · TEMPERATURE
Fleets and mobile assets
Combine position, vehicle telemetry, geofences, and alerts across thousands of tracker models.
WATER · SOIL · WEATHER · TANKS
Environment and agriculture
Operate remote sites where batteries last for years and there is no Wi-Fi or mains power.
The next input layer · Coming soon
Kilo Lens brings video into physical operations
Organize supported RTSP and ONVIF cameras across sites, recordings, motion zones, videowalls, and scoped access — next to the sensors and device state Kilo already understands.
Preview Kilo Lens→Physical AI infrastructure
Physical AI FAQ
Is Kilo a robotics framework?+
Kilo is the governed IoT execution layer behind physical systems. It connects telemetry, automation, commands, permissions, and auditability; a robot or controller can integrate through supported protocols and APIs.
Does an external AI agent need an API key?+
Not through MCP. A compatible client discovers Kilo’s OAuth flow and opens browser sign-in. The connection receives the signed-in user’s permissions. Traditional integrations can still use scoped API keys.
Can an AI directly control a device?+
Only through the permissions and controls configured in Kilo. Commands can use typed parameters, approval gates, audit history, and optional verification; automation can be tested and versioned before deployment.
Does Kilo depend on one model provider?+
No. Use Kilo’s included AI allowance, bring OpenAI, Anthropic, Ollama, or another compatible endpoint, or connect an independent MCP-capable client.
Give your AI a safe way to act
Start with five devices free, connect your AI client, or talk through an enterprise physical-AI architecture.