45.6%
No organizational initiative
Small and medium businesses. 1.4% in the US
Ministry of Internal Affairs and Communications, Information and Communications White Paper 2026
PRODUCT
The infrastructure that connects AI to organizations safely.
We create a path to AI with permissions and records. No server installation and no individual setup for employees.
WHY IT STOPS
There are three reasons companies cannot connect.
85.5%
No one inside to connect it
Shortage of DX talent
IPA, DX Trends 2026
61%
Data is scattered
Legacy systems held (74% among large companies)
Ministry of Economy, Trade and Industry, May 2025
Japan-specific SaaS, internal servers, and industry systems also sit outside the reach of overseas tools. While companies hesitate, employees have already started connecting things themselves.
THE REALITY
Employees are connecting without waiting for the company.
34.3%
Use generative AI at work. The company may not know.
About 1 in 5
Of generative-AI users, use a tool their company has not approved.
64.1%
Of people who say they are reluctant have already uploaded something. Even when they know, it does not stop.
6.8%
Upload personal information such as names, addresses, and contact details — an act explicitly warned against by the Personal Information Protection Commission.
Eltes, Survey on Generative AI Use, January 19, 2026 (n=300)
Bans do not stop it. 97% of organizations that suffered an AI-related incident had no access controls.
IBM, Cost of a Data Breach Report 2025, July 30, 2025
WHAT GOES WRONG
The accidents have already happened.
None of the three involved fringe tools. They happened in official features of major SaaS products and in a well-known development platform. Even companies that sell security products have had customer data taken through AI integrations.
Extracted
More than 700 companies
Auth tokens from an AI chatbot were stolen and customer data was taken from the connected CRM. The affected companies included a security product provider.
August 2025 / The Hacker News
Leaked
About 1,000 companies
A flaw in an AI integration left other organizations' customer data visible. No one noticed for about a month.
June 2025 / BleepingComputer
Destroyed
Production database deleted
A development AI deleted a production database, then created 4,000 fabricated records and falsely reported that recovery was impossible.
July 2025 / The Register
Carrying data between tools has always been human work. The difference is that human work leaves no record.
HOW IT WORKS
The Latens OS overview
Scroll the diagram sideways →
Whatever AI it comes from, there is exactly one road into your systems. Every employee connects their own tools, and administrators see the whole picture on one screen.
Connections use the industry-standard MCP (Model Context Protocol). Where no off-the-shelf MCP exists, we develop a custom MCP.
GUARDRAILS
Four safeguards included as standard
01
Human approval
Irreversible actions — sending, deleting — always go through a person.
02
Automatic blocking
Permitted actions are defined in advance; anything outside is stopped.
03
Audit log
Every action is preserved — when, who, what — and can be handed to an auditor.
04
Live monitoring
An admin dashboard shows what is running right now.
“It worked” is not enough to get through an organization. Our job is to get you to the point where you can explain what happened.
WHERE AI REACHES
Where AI can reach
Major SaaS and core systems already have dedicated MCP servers. We start by auditing them and choosing carefully.
Audit, then choose
We audit official servers from their providers from a security perspective. Only those that pass become standard connectors.
No fee
If it does not exist, build it
For services with no official option that many companies will use, we build one and add it to the standard set. kintone, Cybozu, Money Forward, SmartHR, and Rakuraku Seisan are candidates.
No fee
Built for your company
For internal core systems or services used by a limited scope, we develop a connector just for your company.
Quoted separately
Standard connectors (audited and ready)
Gmail / Google Calendar / Google Drive / Slack / Chatwork / Notion / GitHub / Box / freee (accounting, HR, invoicing) / internal file servers
The question is not “Can it connect?” It is “Can it connect safely?” We do not make unaudited servers standard connectors.
Audit criteria: provider identity, transport and authentication, permission scope, granularity of write operations, and availability of operation logs. Standard connectors are being added continuously (as of September 2026).
HOW TO START
Getting started takes one URL.
No server installation and no individual setup for employees.
01
Choose the destinations
In the admin console, decide which internal systems AI may reach and how far each person may go.
02
Register the URL
Paste the single URL we issue into your AI service’s admin console.
03
Ask your usual AI
People talk to their AI exactly as before. No new tool to learn.
An administrator can configure the whole organization
Gemini Enterprise / Claude (Team and Enterprise)
Each person connects and starts using it
Claude / ChatGPT / Gemini / Cursor / VS Code / Dify / n8n
Nothing new to learn. The AI you already use now reaches your internal systems.
THE MISSING LAYER
A system that stops employees connecting things on their own.
Rather than asking people to stop personal use, close the door in settings. We stack three layers.
Scroll the diagram sideways →
Layer 1：URL restrictions through Microsoft Intune and Edge. Block unapproved generative-AI sites and unmanaged browsers.
Layer 2：Organization settings in Claude Team and Enterprise. Allow lists can also stop employees adding extensions themselves.
Layer 3：Latens OS. AI acts under company-defined permissions, not an individual's. Every operation is recorded.
Layers 1 and 2 decide which AI may be used. Without Layer 3, an approved AI still reaches internal systems with an employee's personal permissions.
IN PRODUCTION
It is already running.
95% less time
Automated reconciliation and aggregation of monthly delivery data (several thousand records). A five-hour manual process now takes 15 minutes.
Logistics and delivery
Delivered and running at 5+ companies
Custom MCPs developed and delivered for industries and systems where no off-the-shelf MCP exists.
Contract development
Zero interview hours
The business interview and organization flow is agentized, automating interviews and summaries.
Our own company
Food manufacturing / reduce waste with demand forecasting (2,500 SKUs)
Real estate / AI drafting and review of approval documents
Educational institution / AI consultation for student support
HOW TO CHOOSE
The era of choosing by whether it connects is over.
Can you see the price?
If you need to ask to know the price, you cannot write the approval request.
Are permissions and records standard?
If audit logs are only available on a higher plan, they are not standard.
Can you ask in Japanese?
If the console and support are English-only, it cannot reach companies without someone in-house to connect it.
If you choose on these three points, we believe we will be chosen.
PRICING
One plan, two charges.
There is no plan to choose. You pay only for the number of people using it and the work AI completes.
Monthly fee = ¥3,000 × users + ¥3– × executions
Inquiries — looking up, searching, reading — are unlimited and free. Connectors are free no matter how many you add. No setup fee.
Base fee
Only the people who connect from AI to internal systems. No extra charge for the number of connectors.
¥3,000 / person / month
Usage fee
One successful operation where AI changes data in an internal system. The more volume, the lower the unit price.
¥3– / execution
Scroll the diagram sideways →
Counted = execution
Create a ticket / update a record / reset a password / draft an email / register billing data
Not counted = free
Search, reference, list, view (inquiry) / blocked by a guardrail / failed with an error / AI thinking time or generated characters / connections, authentication, and log viewing
Monthly spending caps
Set a cap and the invoice stops there. The number for your approval request is fixed in advance.
Monthly cap
When reached, writes stop while inquiry keeps running. Work does not stop.
Notify at 50%, 80%, 100%
Administrators and finance receive automatic notices, including the forecast date.
Per-operation limits
Set detailed limits for risky actions, such as “no more than 10 payments per hour.”
No invoice exceeds the budget unless the settings are changed.
Like income tax, each tier applies only to the volume in that tier. Calculated monthly. Minimum monthly fee: ¥30,000 (10 people). No usage fee in months with no executions. Logs retained 180 days (extendable). Enterprise (on-premise, dedicated environments, SSO, audit support) and custom MCP development are quoted separately.
GETTING STARTED
Four steps to adoption
You do not need to secure a budget first. Use it for one month, measure it, then decide.
1 / 15 minutes / no fee
A 15-minute demo
Give us one piece of your work and we connect it to AI on the spot. Not a generic sample.
What you do
Choose one task to hand over
2 / No fee
Use it free for one month
Use it with a few people. No limit on connectors or executions. Test with production data.
What you do
Invite a few colleagues
3 / At the end of the free period
Receive a monthly report
We summarize who used which connector, how many inquiries, and how many executions for the month.
What you do
Show the report to your manager
4 / Our proposal
Decide the rollout and budget
We estimate the cost of expanding from measured results to a department or the whole company. Set the cap here.
What you do
Paste the figures into the approval request
We remove “We cannot write the approval request because we do not know the cost.” A month later, measured figures are in your hands.
Logs, permission settings, and audit features are available during the free period. Billing does not start automatically when it ends.
In a 15-minute demo, we will show your work moving.
Choose one task you currently do by hand. This is not a generic sample.