Meet us at Welcome Tbilisi Tbilisi
AI that sees, reads and gets the work done
Our products use the cameras you already have to watch lighting and building sites, and turn client requests into 1C and SAP documents. Whatever else you need, we build it – from pilot to production in weeks, not years.
-
~ 1.5 s
to detect a dead light on camera
LightControl
-
~ 2 min
from a client’s request to a draft offer in 1C
T-BOX
-
2 – 6 weeks
for a pilot on your own data, with a measured result
Pilot
-
3 languages
we work in Georgian, Russian and English
ka · ru · en
The stack we ship with
Products
Five products. One team that ships.
Our own AI products, built and supported by the same engineers who will take on your project. Pick one to see it at work.
-
LightControl
AI lighting monitoring
Finds dead, dimmed and flickering lights through the CCTV you already have, and tells the maintenance team exactly which one.
~1.5 s to detect a failed light
-
DangerZone
AI site safety
Watches people, machines and no-go zones on site and alerts the site manager within seconds.
5 types of violations recognised
-
T-BOX
AI automation for 1C & SAP
Turns a client’s Excel, PDF or email request into a draft offer in 1C:ERP or SAP.
~2 min from request to draft offer
-
AI Gateway
Private LLM access with data masking
Masks personal data before a prompt reaches ChatGPT or Claude, and puts it back into the answer.
Masked before a prompt reaches the LLM
-
HarvestEye
AI quality-control device
Four cameras measure every fruit’s calibre, find defects and assign a grade. The report writes itself.
4 cameras: calibre, defects, grade
-
Custom AI & software
And anything else
LLM agents, computer vision, 1C/SAP, web and mobile apps, IoT. You bring the task, we bring the team.
2–6 wk to a working pilot
AI lighting monitoring
LightControl spots a failed light before anyone else does
It watches architectural and landscape lighting through the CCTV cameras you already have. No sensors on the fixtures, no night patrols: within seconds, the right team knows exactly which fixture has failed.
-
Works on your existing CCTV
At night, with ordinary IP cameras. Nothing to mount on the fixtures.
-
Catches five kinds of failure
Off, dimmed, wrong colour, flickering, or a whole circuit gone dark.
-
Speaks your maintenance team’s language
“Bollard B-16 off · garden courtyard · circuit Y4” arrives in Telegram, email or your helpdesk.
-
Closes the incident by itself
Once the camera sees the light back on, the incident closes and the before-and-after photos stay as proof.
- ~1.5 s to detect
- ~3 s to notify
- auto close after repair
Measured on our demo site: 8 cameras, 320 fixtures
Interactive demo
Break a light. Tap any light in the scene or pick a fault below, then watch the AI find it.
Site health
- 320 / 320fixtures OK
- 8 / 8cameras online
- Open incidents: 0
Active incidents 0
No open incidents
All lights in view are working. Break one and see what happens.
Event log
- INFOPatrol mode · 17 fixtures under watch
The real product
Not a mock-up. Screens from our demo site: 8 cameras, 320 fixtures, incidents opened and closed by the AI.
DangerZone · AI site safety
Sees the danger before the accident
One fixed camera watches people, machines and restricted zones on a construction site, on a factory floor or in a warehouse. DangerZone flags each violation as it happens and shows who broke which rule, and where.
What it flags
- Person in a no-go zone (Alarm) Excavation pits, crane load areas, electrical rooms
- Machine in a no-machine zone (Alarm) Excavators and forklifts on pedestrian routes
- Worker near a moving machine (Alarm) Inside the danger radius while it drives or swings
- No hard hat (Warning) Checked for every person in view, all shift long
- Running where it is not allowed (Warning) Scaffolding, stairs, busy aisles
Privacy built in
Faces are blurred. Public areas beyond the fence are ignored. Exempt roles, such as the operator in the cab, never raise an alert.
- Alerts reach the site manager in seconds
- Reports for HSE audits: snapshots, timeline, replay
Pilot on your own cameras in 2–6 weeks
Alerts 0
On site 5 workers
Hard hats 4/5
Drag a worker into the red zone or next to the excavator
Violations Newest first
All clear for now
Drag a worker into the red zone or next to the excavator. With a keyboard: Tab to a worker, then move it with the arrow keys.
Product screenshot
Not a mock-up: the real console
T-BOX · AI for 1C and SAP
From a client’s request to a 1C offer in two minutes
Requests arrive as Excel sheets, PDFs and hurried emails. T-BOX reads them, finds every item in your catalogue and drafts a commercial offer in 1C or SAP. The manager checks it and sends it. Nobody retypes a thing.
- ~2 min from a request to a draft offer
- 0 invented items: rules pick every SKU
- 1C + SAP one approach for 1C:ERP, 1C:CRM, SAP S/4HANA and Business One
Client request
#DescriptionQtyUnit
1 Pipe 57x3 AISI 304 120 m
2 Elbow 90° 57×3 40 pcs
3 Tee 57*3 12 pcs
4 Reducer conc. 57x3–38x2 10 pcs.
5 Flange DN50 PN16 16 pcs
6 Pipe 38x2 AISI304 60 lin. m
7 Ball valve DN50 6 pcs
Delivery: Batumi, by 30 Oct
Your system
Processing time
00:00
- Extract LLM reads the request —
- Match rules pick the SKU —
- Prices & stock from 1C:ERP —
- Draft document offer + manager task —
# Client wording Qty SKU Match Price Stock
1
2
3
4
5
6
7
Total —
Press “Process” and T-BOX reads the request line by line.
Commercial offer
7 lines 18,450 ₾ 1 line to confirm
- request_14-10.xlsx · original file attached
- Task for the manager: confirm line 7 and send the offer
How T-BOX works
-
An LLM reads the request
Excel, Word, PDF or a plain email: lines, quantities and units are extracted, even with typos and mixed units.
-
Rules pick the SKU, not the LLM
Matching to your catalogue is deterministic, so no invented items. Doubtful lines are flagged for the manager.
-
Prices and stock from 1C or SAP
Current price lists, discounts and warehouse balances, straight from your system.
-
The client’s wording stays
Each line keeps the client’s own name for the item next to your SKU, so nothing gets lost.
-
The original is attached
The source file goes into the document, ready for any later check.
-
The manager gets a task
A draft offer and a task appear in 1C or SAP. The manager reviews, adjusts and sends.
Works with
- 1C:ERP
- 1C:CRM
- SAP S/4HANA
- SAP Business One
Beyond request intake
- Invoice and document recognition
- Reconciliation
- RS.ge exchange
- Reports
- AI assistant inside 1C and SAP
Run T-BOX on your requests
Pilot on your 1C or SAP in 2–6 weeks
AI Gateway
Use ChatGPT and Claude without leaking personal data
AI Gateway sits between your team and the LLM providers. It finds personal data in every prompt, swaps it for placeholders, sends only the masked text and puts the originals back into the answer. People keep using AI; customer data stays inside the company.
Policy Tap a category to switch: mask → allow → block
Your employee wrote
Edit the text: add your own name, phone or card number.
What the LLM receives
Gateway finds and masks
Answer your employee sees
LLM answers, gateway restores
0 masked
0 personal data left the company
request blocked by policy
Demo: detection runs in your browser and the answer is a template. Nothing is sent anywhere.
How it works
-
Detect
Names, personal numbers, phone numbers, IBANs, card numbers, contracts and credentials, in every prompt.
-
Mask
Each value becomes a placeholder like [NAME_1]. The same value always gets the same placeholder, so the answer still makes sense.
-
Send
Only the masked request goes to OpenAI, Anthropic or your own model.
-
Restore
The originals go back into the answer while it streams. The employee reads a normal, complete reply.
Policies per category
Mask, allow or block each type of data, separately for teams and systems.
Audit without content
Every request is logged: who, when, which model, what was masked. Never the text itself.
Drop-in API and web chat
OpenAI- and Anthropic-compatible API for your systems, a web chat for staff and a mask/unmask API.
Fails closed
If a check cannot run, the request does not leave. No content is stored.
On-premise if you need it
Runs inside your network next to your data, or in your private cloud.
Evidence for compliance
Policies and the audit trail help you demonstrate compliance with the Law of Georgia on Personal Data Protection and GDPR.
Try AI Gateway in your company
Pilot on your data in 2–6 weeks
AI gadgets
HarvestEye grades a sample in 40 seconds
A four-camera device for fruit and vegetable quality control. Neural networks find every item in the sample, measure its calibre, detect defects and assign a quality grade. From task to accepted report, with no paper forms.
Built for
- Importers and exporters
- Distribution centres
- Retail chains
- Farms and packing houses
12 items on the tray · ready to inspect
- Find
- Measure
- Defects
- Grade
- Report
- CAM 1 12
- CAM 2 12
- CAM 3 12
- CAM 4 12
Inspection report
- Items
- —
- Avg calibre
- —mm
- Defect rate
- —%
- Batch grade
- —
Time to report
HarvestEye ~40 s
Manual check, typical ~40 min
Simulated in your browser. In our tests, the real device inspects a sample in about 40 seconds.
The device is live at our booth in Tbilisi. Come and watch it work, or let’s plan a pilot on your own produce.
Other AI gadgets we build
When AI has to run on site rather than in the cloud, we design the device around the task.
-
Smart cameras with on-device AI
Detection and counting run inside the camera: video stays on site, only events leave it.
-
Kiosks and voice assistants
Self-service kiosks and voice assistants for lobbies, shops and service desks, in Georgian, Russian and English.
-
Sensors and vision on Jetson and ARM
Cameras and sensors on NVIDIA Jetson and ARM boards for production lines, warehouses and greenhouses. They keep working offline.
-
Hardware, from prototype to small batch
Electronics, enclosure, firmware and the software around it, from the first working unit to a small production run.
Services
Anything else? We build it.
Our products cover the common cases. For everything else, there is our engineering team: AI, computer vision, ERP and software, built end to end. Fixed price per stage, a demo every week, and the code and data are yours.
What do you need?
Pick a task and we’ll highlight the service that covers it.
AI integration
Assistants and agents that draw on your documents and work inside your systems.
- LLM assistants and agents
- RAG search over your documents
- Telegram, WhatsApp and web chatbots
- Voice bots and on-premise LLMs
Computer vision
We teach your existing cameras to count, read and check quality.
- Custom detection and counting
- OCR for documents and labels
- Quality control on the line
- Video analytics on existing cameras and edge devices
ERP automation
1C and SAP without retyping: documents, reconciliations and data exchange run by themselves.
- 1C: ERP, Trade Management, Accounting
- SAP S/4HANA, Business One, Odoo
- Banks, marketplaces and EDI
- Data exchange with RS.ge
Software development
Web platforms, mobile apps and backends, from design to launch and support, by one team.
- Web platforms and dashboards
- iOS and Android apps
- Backends and APIs
- Cloud, DevOps, legacy modernisation
AI gadgets & IoT
Devices that see and decide on site, from prototype to small-batch production.
- Smart cameras with on-device AI
- Kiosks and voice assistants
- Edge AI on Jetson and ARM
- Prototypes and small batches
Consulting & audits
A clear plan before the budget: where AI pays off and what it takes.
- AI strategy and roadmap
- Audit of automation potential
- Data readiness review
- Security and compliance
Not on the list?
Describe the task. We’ll suggest an approach, a team and a first estimate.
Industries
Where AI pays off first
Pick your industry to see three use cases we can start with, and which of our products cover them.
Construction
Big sites, moving machines and paperwork that never stops.
-
Danger zones on camera
People in no-go zones, workers next to moving machinery, missing hard hats: the site manager gets an alert with a snapshot in seconds.
-
Gate and delivery control
Cameras count trucks, deliveries and equipment at the gate and compare them with the plan.
-
Requests straight into 1C
Requests and specifications in Excel or PDF become draft 1C documents in about two minutes, with prices and stock from 1C.
Real estate & facility management
Façades, courtyards and tenants who expect everything to just work.
-
Façade and landscape lighting
Existing CCTV finds dead, dimmed and flickering fixtures, notifies the maintenance team and closes the incident once a camera sees the repair.
-
Tenant requests in a chat
A Telegram or WhatsApp bot takes requests in three languages, answers routine questions and files tickets in your helpdesk.
-
Proof for owners and SLAs
Before-and-after photos of every repair, a health score per building and view-only links for owners and clients.
Manufacturing
Lines, forklifts and quality that has to be proven.
-
Safety on the shop floor
Restricted zones around machines and forklift lanes. Faces are blurred and reports are ready for HSE audits.
-
Visual quality control
Cameras on the line catch defects, count output and read labels before goods leave the plant.
-
Orders and invoices without retyping
Customer orders and supplier invoices go straight into 1C or SAP; catalogue matching follows strict rules.
Retail & distribution
Thousands of SKUs, hundreds of requests and fresh goods at the gate.
-
From request to draft offer in two minutes
A customer’s Excel, PDF or email becomes a draft offer in 1C in about two minutes, with live prices and stock.
-
Produce acceptance
HarvestEye grades incoming fruit and vegetables by calibre and defects before you sign the delivery note.
-
ChatGPT without data leaks
AI Gateway masks customer names, phone numbers and card data before any request reaches an LLM.
Logistics & warehouses
Forklifts, pallets and documents that must match the goods.
-
People and forklifts kept apart
An alert when a person walks into a forklift zone or a machine enters a pedestrian area.
-
Counting and reading on camera
Cameras count pallets and read labels and container numbers right at the dock.
-
Documents that post themselves
Waybills and invoices are recognised and posted to 1C or SAP, with exchange to marketplaces and EDI partners.
Agriculture & food
Citrus, hazelnuts, grapes and stone fruit: quality sets the price.
-
Grading at intake
HarvestEye measures calibre, finds defects and grades a sample in about 40 seconds, replacing the manual check.
-
Export paperwork in order
Lots, stock and export documents are kept in 1C and exchanged with RS.ge automatically.
-
Sensors in storage and greenhouses
Edge devices watch temperature, humidity and ripeness, and warn you while the produce can still be saved.
Hospitality & HoReCa
Guests in three languages and a façade that works for the brand at night.
-
A guest assistant
Bookings, answers to common questions and requests to staff in Telegram, WhatsApp or on the website, in Georgian, Russian and English.
-
Lighting checked every night
Façade and garden lighting checked every night; the team learns about a failure within seconds.
-
Purchasing on autopilot
Supplier invoices are recognised and posted to 1C; stock and purchase orders stay up to date.
Finance, public sector & smart city
Strict rules for personal data, and whole cities to keep lit.
-
LLMs under compliance control
AI Gateway masks personal IDs, IBANs and names before a request leaves the bank and audits every call, helping you meet Georgia’s personal data law and GDPR.
-
Streets, bridges and parks
City CCTV checks street, bridge and park lighting every night and sends failures straight to the contractor.
-
Assistants for regulations
RAG assistants answer staff questions from internal regulations and documents, running on on-premise LLMs.
Calculator
See what AI saves you – in real time
Move the sliders and watch hours, money and payback recalculate. Switch to “Project estimate” for an indicative budget, timeline and team in under a minute.
01Scenario
02The work today
03With AI
Typical for this scenario: $18,000
Plus support and hosting at $500 a month, already included in payback and ROI.
01Modules
Tick what you need and set the scale. Any combination works.
Platform
e.g. requests → offers, invoices, delivery notes
Size
Several sources, CRM or helpdesk integration
Size
Web and mobile, integrations
02Options
Deployment
Support
Indicative figures, VAT not included. We confirm scope, timeline and price after a free discovery call.
Why us
Why teams hire us
We bring working software instead of presentations, prove the result on your data and hand you the keys.
- 5
AI products of our own
Products already running, not slides
LightControl, DangerZone, T-BOX, AI Gateway and HarvestEye. Try each of them live at our booth.
- 2–6
weeks to a working pilot
A pilot on your data
A working prototype on your cameras, documents or 1C, with a measured result instead of a promise.
- 7
days between demos
Fixed price, a demo every week
The price of each stage is agreed before it starts. Progress is on your screen every week, not in a report.
- All
code, docs and access are yours
You own the code and the data
At the production stage we hand over the repository, documentation and admin access. No vendor lock-in.
- GDPR
and Georgian data-protection law, built in by design
On-premise or cloud, security first
Your servers or the cloud you choose. Personal data is masked and access is logged.
- 3
working languages
We speak Georgian, Russian and English
ქართული · Русский · English, in meetings, documentation and interfaces.
The usual way vs with us
The same project, two ways to run it. Flip the switch.
Time to the first working result
months 2–6 weeks
- Start Months of specifications A working pilot in 2–6 weeks
- Proof Slides and promises A live demo on your cameras and data
- Budget Hourly billing with no ceiling A fixed price for every stage
- Progress Silence until the deadline A demo every week
- Result A black box you rent Your code, your data, your infrastructure
How we work
From first meeting to production in weeks
Four stages, each with a fixed price and a clear result. After every stage, you decide whether to continue.
-
Stage 01
Discovery
Free 1–2 meetings
We learn your goals, data and systems and agree on how success will be measured.
What you get
- Goals and KPIs on one page
- A review of your data, cameras and systems
- A pilot plan with a fixed price
-
Stage 02
Pilot
2–6 weeks
A working prototype on your data or your cameras, with a measured result.
What you get
- A prototype running in your environment
- Results measured against the KPIs
- A demo every week
- A production plan with a fixed price
-
Stage 03
Production
4–12 weeks
We integrate, secure and roll out the system and train your team to run it.
What you get
- Integration with 1C, SAP, cameras and helpdesk
- Security, access rights and audit
- Training and documentation
- Source code and access handed over
-
Stage 04
Support & growth
SLA Ongoing
We keep the system healthy and grow it with new use cases.
What you get
- Monitoring under an SLA
- Model updates as your data changes
- New use cases on the same foundation
- Fixed price per stage
- A demo every week
- Your code and data
- On-premise or cloud
Conference
Meet us in Tbilisi
Five live demos on real data, and the engineers who built them. Come to our booth or book a 20-minute meeting in advance.
Starts in
-- days
-- hours
-- min
-- sec
Happening now
We are at the booth today. Come by or book a slot below.
Live
- Event
- Welcome Tbilisi
- Dates
- October 23 – 25, 2026
- Venue
- Tbilisi City Assembly Building, 2 Freedom Square
What you will see at our booth
Every demo runs live. Bring your own file or question.
- Station 01
LightControl camera wall
A live wall of cameras: the AI spots a failed light in about 1.5 seconds and tells the team exactly which one.
Try it online - Station 02
DangerZone replay
A real construction-site clip: people, machines and no-go zones, with violations flagged in seconds.
Try it online - Station 03
T-BOX: Excel → 1C
Hand us an Excel request and watch it turn into a draft offer in 1C in about two minutes.
Try it online - Station 04
AI Gateway, live masking
Type a name, an ID or an IBAN: it is masked before the LLM sees it and restored in the answer.
Try it online - Station 05
HarvestEye device
Four cameras grade fruit on the spot: calibre, defects and quality class.
Try it online
Meeting booker
Book a 20-minute meeting
Pick a day, a time and the topics. We will confirm the meeting by email or phone.
Welcome Tbilisi
Tbilisi · 2026
Day Friday, October 23
Time 20 min
Topics
This is a request, not a confirmed booking. We reply to every one.
FAQ
Questions we hear at every meeting
Short, concrete answers. If yours is not here, ask us directly.
Have another question?
Send it through the form or ask us at our booth in Tbilisi.
Do you work outside Georgia? In which languages?
Yes. We work with teams in Georgia, the Caucasus and Central Asia, and take on projects in other countries too. Most of the work runs remotely, and we come on site when the project needs it. Meetings, documentation and interfaces are in Georgian, Russian or English.
Can it run on our own servers?
Yes. Our products and custom systems can run on your servers, in your private cloud or in a public cloud of your choice. Language models can also run inside your network, so the data never leaves it.
Which cameras work with LightControl and DangerZone?
Most IP cameras that stream over RTSP or support ONVIF, the standard protocols of modern CCTV. As a rule, no new cameras or sensors are needed: during the pilot we connect to yours and check image quality, angles and performance at night.
How long does a pilot take, and how much does it cost?
Usually 2 to 6 weeks, depending on the task and the data. The price of each stage is fixed before work starts, so you never pay by the hour. The project calculator gives an indicative range, and discovery meetings are free.
Which 1C and SAP versions do you support?
1C:ERP, 1C:CRM, Trade Management and Accounting on the 1C:Enterprise 8.3 platform; SAP S/4HANA and SAP Business One through their APIs. We also work with Odoo and set up data exchange with RS.ge and banks. If your configuration is customised, we review it during discovery.
Is our data used to train models?
No. Your data is used only in your project and never to train models for anyone else. With AI Gateway, LLM providers receive only masked requests. Models we fine-tune on your data belong to you.
How do you comply with personal data laws?
We design for the Law of Georgia on Personal Data Protection and for GDPR from the start: only the data the task needs, access control and audit logs, faces blurred in DangerZone, personal data masked before it reaches an LLM, and storage on your servers when required.
What support do we get after launch?
Support runs under an SLA: monitoring, fixes, model updates as your data changes, and new use cases. Response times and support hours are set in the contract to match how critical the system is for you.
Will you sign an NDA?
Yes, before we see any of your data or documents. We can sign your template or offer ours.
How do we start?
Send a request through the form or book a meeting at the conference. The first one or two meetings are free: we agree on goals, look at your data and systems, and come back with a pilot plan and a fixed price.
Start a project
Tell us what to automate
A process, a camera, a 1C database or just an idea: describe it in a few lines. An engineer reads every request and replies with clarifying questions or a first plan.
- We aim to reply within one business day Your request goes to an engineer, not a call centre.
- Georgian, Russian or English We reply in the language you choose in the form.
- NDA on request We sign it before you share any details.
- Request number
What happens next
-
We read and reply
~1 dayClarifying questions or a first idea of the solution.
-
Free discovery meeting
freeOne or two meetings, online or in person: goals, data, systems and how we will measure success.
-
A pilot plan with a fixed price
pilot 2–6 weeksScope, timeline and a fixed price for each stage. You decide whether to go ahead.