LITA runs the day-to-day of your site: one action model, and the operating system built around it, connected to your software, sensors and robots. You set the intent. It spots what needs doing, turns it into tasks, hands them to an agent, a robot or a person, and checks the work.
When a site's software, sensors and robots all report to LITA, it keeps a live 3D model of the operation: every room, machine, robot and person is a node in one graph, with a position, a status and what it is linked to. Agents query that graph the way you'd query a database. So can you.
Choose a site, then watch LITA detect, decide, coordinate and verify. All scenarios, names and figures are illustrative. Query the graph to explore, or drag the model to turn it.
One person on shift. Two cameras, a POS, fridge and shelf sensors, smart lockers and a sidewalk delivery robot.
LITA
Station kiosk1 site
IntentNever run out of the top 20 items. Keep the queue under 3.Autonomy: act under USD 100 Live 07:41
Ordered 24 from the dairy supplier for USD 38.40, under your USD 100 limit, and booked the delivery robot for 07:52.
A production line, CNC cells, robot arms, mobile robots and a loading dock, tied to the plant's MES and ERP.
LITA
Plant 2 · AssemblyLine 1, cells A to C
IntentHit 6,500 units a day with no unplanned stops on Line 1.Autonomy: ask first on stops Live 09:16
Put the gripper swap into the 14:00 changeover so Line 1 doesn't stop. Shift lead approved at 09:14.
Ten rooms over a lobby and restaurant. Door and occupancy sensors, the booking system, six staff and four robots on one floor plan.
LITA
Linden Court Hotel3 floors · 10 rooms
IntentEvery room ready by check-in. No heating in empty rooms.Autonomy: act on rooms and energy Live 11:05
Moved 203 ahead of 204 for the early arrival and messaged the guest a ready time.
You set up the intent once. After that, LITA runs a continuous loop around a live state of your business: observe, detect, plan, assign, orchestrate, check. Click any step to see its screen.
Set up
Understand
Act
Check
Grow
State what the operation is for and the rules it must never break. That is the only required input. LITA asks for the rest.
LITA reads the intent, spots what it would have to guess, and asks. Answers become part of the operation's rules.
Each system connects through an MCP server: your booking system, messaging, locks, sensors, cameras, elevators and robot fleet. LITA reads what each one can report and change, and treats them all as one set of hands. High-rate feeds like cameras and sensors stream into the live state as events; MCP carries the queries and the actions.
| MCP server | Kind | What LITA gets | Status |
|---|---|---|---|
| pms.harborlane | software | bookings, room status, 14 tools | connected |
| whatsapp-business | software | staff and guest messages | connected |
| sms-voice | software | SMS and phone calls | connected |
| keycard-locks | iot | 124 locks, time-limited unlock | connected |
| door-contacts | iot | 124 sensors | connected |
| occupancy | iot | 124 sensors | connected |
| thermostats | iot | 124 setpoints | connected |
| corridor-cams | camera | 8 feeds, vision on corridors only | connected |
| elevator-ctl | iot | 2 cars, call and hold | connected |
| fleet | robots | vacuum-v01, vacuum-v02, linen-l01 | connected |
Every room, device, robot and person becomes a node with a position and a current status. For a facility, the state is a 3D graph: floors, rooms, corridors, elevators. Agents query it the way you'd query a database.
Signals come in from camera vision, sensor data like occupancy and movement rate, events in open software, and push messages. Raw feeds become events before they reach the loop; LITA separates incidents (something is wrong) from requests (someone wants something).
Two signals about room 214 merge into one task with a deadline, subtasks in order, the skills and access each needs, and what "done" means, taken from your checklist.
For each subtask LITA checks skill, location, availability and, for robots, whether the battery covers the trip, the job and the return. People get it by SMS, call or ping.
| Subtask | Candidate | Skill | Battery or time | Result |
|---|---|---|---|---|
| 2 Vacuum 214 | vacuum-v01 | yes | 34%, needs 41% | skipped |
| 2 Vacuum 214 | vacuum-v02 | yes | 82%, 3 min away | assigned |
| 1 Linen | linen-l01 | yes | 58% | assigned |
| 3 Room | Ana | yes | busy until 11:40 | backup |
| 3 Room | Luis | yes | floor 2, free | assigned |
| 4 Inspect | qa-inspector | yes | available | assigned |
A robot can't vacuum a room it can't reach. LITA calls the elevator, holds the doors, unlocks the room for a limited time, moves the robot into position, runs the right skill, and locks up after.
Each part of the task is checked against its own evidence: robot coverage maps, photos from staff checked against the checklist, and system status. No cameras inside guest rooms.
Failed checks go back to whoever can fix them fastest. Passed work closes, updates the state, and triggers what depends on it.
Once one role runs well, LITA proposes the next, based on what it's already seeing. Each role is the same loop with its own rules and limits. You switch them on one at a time.
LITA hosts the action model, the agents that run on it, the state and the connections. The console shows every task, who has it, what the robots are doing, and what's waiting for you.
| Waiting for you | Role | Why it asks |
|---|---|---|
| Overtime for Ana, 1 h Saturday | Housekeeping | overtime rule |
| Linen order, USD 780 | Finance | over USD 500 |
| Enable Kitchen role | General | new role |
Before it does anything, LITA takes inventory, places every device in the building, and learns what a normal day looks like. Its first output is findings, not actions.
Reads every MCP server's tools and resources: what it can see, what it can change.
Finds every device and matches it to a place. Anything it can't place, it asks you about.
Builds the live state and, for a facility, the 3D graph from floor plans and device positions.
Learns normal: occupancy by hour, how long jobs take, where the bottlenecks are.
Surfaces problems it can already see, as suggestions. You decide what it may act on.
You don't need to wire up the whole shop on day one. LITA works out what it can do with what you have, tells you what the next device would unlock, and asks for more authority on the actions it keeps getting right.
Unlocks with more connected
Humans, robots and agents on one board. LITA hosts it, so there is nothing to install on site beyond the devices you choose to connect.
| Name | Type | Now | Battery or shift | Reached by |
|---|---|---|---|---|
| Ana | person | rooms 317 to 322 | shift to 16:00 | WhatsApp, call |
| Luis | person | floor 2 turns | shift to 16:00 | WhatsApp, call |
| Maria | approver | 3 approvals waiting | front office | ping |
| vacuum-v01 | robot | docked, charging | fleet | |
| vacuum-v02 | robot | floor 3 corridor | fleet | |
| linen-l01 | robot | to room 118 | fleet | |
| front-desk | agent | 3 guest threads open | always on | in LITA |
| qa-inspector | agent | room 214 re-check | always on | in LITA |
| energy | agent | 11 setbacks running | always on | in LITA |
Starting points, not limits. Each pairs an intent you'd type with the software, sensors and robots it would use. If your business has a building, stock or machines, it fits.
Reads checkouts and door contacts, routes housekeepers, sends cleaning and linen robots, sets room temperature by occupancy, answers guest messages.
Watches sales and fridge levels, places supplier orders, adjusts tomorrow's bake list against the forecast, and flags a fridge drifting warm before stock spoils.
Slots incoming stock by demand, dispatches picking robots, balances pickers across shifts, and reorders from sales velocity instead of a fixed weekly list.
Handles member access, runs ventilation and lighting by headcount, notices machines nobody uses anymore and books a technician, sends the cleaning robot after peak.
Monitors each machine's power draw and error codes, schedules maintenance before a breakdown, runs time-of-day pricing, and handles pickup and drop-off orders.
Runs listings and pricing by occupancy, issues digital keys, sends payment reminders, schedules patrol robots, and opens a ticket when a camera or sensor sees something off.
Every system connects through MCP; live feeds from cameras and sensors stream into the state alongside it. Bring your own servers, or use ours. People stay in the loop through the tools they already use.
A wrong line of code fails a test. A wrong action in a building spends money or moves a machine near people. LITA is built around that difference.
Observe, suggest, ask first, or act. Set per action type, raised one at a time.
Caps per order, per day and per supplier. Anything above waits for you.
Robots keep to the zones and hours you set. A device's own safety stops sit below LITA and can't be overridden by it or by any agent.
Every change shows what the agent saw, what it did and why, like a commit history for the business.
Revert a schedule, a price or a setpoint in one step. Anything irreversible always asks first.
Choose where vision runs. Guest rooms, changing rooms and restrooms stay out.
We're onboarding a small number of operators first. Tell us what you run, what's already connected, and what you'd hand over first.
As autonomous as you set it. Every action type starts at suggest. Most operators move routine work like robot runs and small reorders to act first, and keep spend and staffing on ask first.
No. Software alone works, and one camera is enough to start. Sensors make the picture accurate. Robots let LITA do physical work without a person.
No. You don't draw flows. You state an intent and rules, and the action model decides the steps from the live state of the site.
Both, and they are kept separate. The core action model is LITA's own, built for physical operations: it perceives, plans and decides. Where another model does a job better, such as language or a vendor's robot control, LITA routes to it. The operating system around all of this is the loop, the live state, the connections, the permissions and the log: it decides what runs, when, and with what authority. Robots do the physical work through their own skills, which LITA calls.
LITA is designed to connect through MCP servers: your own, a vendor’s, or ours. Tell us what you use when you join the waitlist so we can discuss compatibility during onboarding.
You see the action and the reasoning in the log, undo it where possible, and lower the autonomy for that action type. Quality checks catch most misses before you do.
Pricing will be shared during early-access onboarding. Join the waitlist and tell us about your sites so we can discuss your requirements.