Add generative AI to existing systems
Replace metered cloud AI with internal AI
Both through Sovereign GaiXer ’s single API gateway
On-premises generative AI product Sovereign GaiXer provides one API gateway for internal systems. Integrate it to add generative AI capabilities to systems that did not previously have them. The same gateway can move workloads from metered cloud AI to a one-time-purchase internal platform. This article starts by explaining APIs, tokens and metered billing in plain language.
In a sentence
Sovereign GaiXer’s API gateway is the shared entry point through which internal systems and apps request work from the unit’s AI. It has two uses: adding generative AI to existing systems, and moving work from metered cloud AI to the internal unit. Both use the same gateway, with application integration and compatibility checks as needed. The Japanese one-time purchase price is JPY 3,806,400 before tax, without extra generation-based charges, excluding operating expenses. Overseas terms require confirmation.
- What you will learn, 1: what APIs, tokens and metered billing mean, using reception desks and order counters as analogies.
- What you will learn, 2: how to add AI to existing systems and move paid cloud workloads to internal processing.
- What you will learn, 3: why one shared entry point matters, and three steps for getting started.
What is an API? An order counter for AI work
First, the terminology: an API is a defined interface through which systems ask each other to perform work. Instead of a person operating a screen, a system asks for a summary or answer in the background and receives the result.
Picture a cafeteria counter. Anyone can order a meal in the agreed format, and the kitchen prepares it. The person ordering does not need to know how the kitchen works. That defined order counter is an API. Sovereign GaiXer’s gateway becomes the counter where internal systems request AI work.
Part 1: Sovereign GaiXer explained describes how Sovereign GaiXer has four layers: PGX hardware, AI-OS, GaiXer App Store and business apps. The API gateway sits in the AI-OS, the third layer from the top. On a smartphone, this is like a map app asking the OS for the current location.
SmartphoneThe device you use every day
Sovereign GaiXer Your company’s generative AI device
A phone OS provides location to maps and camera access to social apps. Apps do not build GPS or cameras themselves. The AI-OS API works similarly: internal systems use AI without building it themselves.
Generative AI (LLM): AI that reads and produces text
Answering questions, summarizing, drafting and classifying are examples of language tasks handled by generative AI. An LLM, or large language model, provides the underlying capability. Sovereign GaiXer hosts that model inside the unit.
Token: a unit AI uses to count text
AI divides text into small units called tokens. The Japanese source uses about 1.5 tokens per Japanese character as an illustrative estimate. Reading 1,000 Japanese characters and producing 500 would then total about 2,250 tokens. Actual counts vary by model, tokenizer and content; this ratio is not an English-text rule.
Metered billing: paying for the tokens used
Many cloud AI APIs charge by tokens. More users and more automated system calls increase usage costs. This article calls rapid growth in these costs a “token cost explosion.”
Cloud AI: external servers. Internal AI: a unit within the company
Cloud AI runs on another provider’s internet-connected servers, often with usage charges. Internal, on-premises or local AI runs on hardware in your company. Sovereign GaiXer follows the latter approach.
Can you add AI to an existing system? Integrate without rebuilding
Yes. The existing system can stay in place. Think of adding a summarize, draft or answer button to sales management, inquiry handling, an internal portal or daily reports. The system asks Sovereign GaiXer’s API to perform the task and displays the returned answer.
Summarize a long inquiry in three lines
When an inquiry is registered, the system automatically requests a summary. Staff use it to prioritize work. They need not consciously operate a separate AI tool.
Answer “What is the travel-expense deadline?” from the policy
Turn a portal search box into a question-answering interface. Retrieval from documents stored in the unit—RAG—is a standard Sovereign GaiXer capability.
Draft a report from bullet-point notes
A salesperson enters a few lines of notes and receives a draft in report format, reducing writing time and standardizing the format for readers.
Read purchase orders and flag possible entry errors
Compare a customer’s purchase-order text with the system record. AI can flag a possible discrepancy such as a quantity that differs between the two.
A smartphone social app does not build its own camera; it uses the OS capability. An internal system likewise need not contain its own AI—it can request it through the API gateway. That avoids rebuilding the whole system.
Can you reduce cloud AI usage charges? Redirect suitable work internally
Systems already calling cloud AI can redirect compatible requests to Sovereign GaiXer’s gateway. It supports cloud-style calls, so the requesting system can use an internal unit instead of an external server. Validate supported API features, output quality and workload performance for the particular integration.
Systems, not just people, call AI
A person using a chat screen can only make so many requests in a day. Once AI is integrated, every inquiry or order can trigger an automatic call. Request volume may become many times greater. That is the mechanism behind rapid token-cost growth.
Usage does not create additional metered fees
The Japanese Sovereign GaiXer offer is a one-time JPY 3,806,400 purchase before tax. More generations or users do not add usage charges, excluding electricity and other operating expenses. Over three years, the purchase price averages approximately JPY 105,733 per month before tax; this is an allocation, not a subscription price. Overseas terms require confirmation.
A mixed approach is practical
Use cloud AI for advanced analysis of public information and Sovereign GaiXer for confidential, high-volume, repetitive tasks. Moving frequent routine work internally can already change how the bill grows.
| Comparison | Cloud AI (metered) | Sovereign GaiXer API gateway |
|---|---|---|
| Pricing basis | Monthly cost varies with token usage | Japanese one-time price: JPY 3,806,400 before tax; no per-use charge, excluding operating expenses |
| As system calls increase | Usage charges increase | No extra usage fee; add capacity if required |
| Data destination | External provider servers | Local processing inside the unit |
| Budget planning | Annual cost depends on actual usage | Hardware purchase up front; operating costs and options remain |
| System changes | — | Redirect compatible calls to the unit and validate integration |
Cloud AI (metered)
- Pricing basis
- Monthly cost varies with token usage
- As system calls increase
- Usage charges increase
- Data destination
- External provider servers
- Budget planning
- Annual cost depends on actual usage
- System changes
- —
Sovereign GaiXer API gateway
- Pricing basis
- Japanese one-time price: JPY 3,806,400 before tax; no per-use charge, excluding operating expenses
- As system calls increase
- No extra usage fee; add capacity if required
- Data destination
- Local processing inside the unit
- Budget planning
- Hardware purchase up front; operating costs and options remain
- System changes
- Redirect compatible calls to the unit and validate integration
Cloud assumptions describe general metered AI services, not a comparison with specific products. Japanese prices are as of September 2026, before tax; overseas terms require confirmation.
It resembles moving work from a pay-per-page copy shop to an office-owned multifunction printer. A small per-page charge seems minor until a system prints thousands of pages daily. With owned equipment, the equipment purchase does not increase per copy, although paper, electricity and other operating costs remain.
Why does one gateway matter? One entry point centralizes management
There are two uses but one gateway. Systems gaining AI and systems replacing cloud calls connect to the same entry point. This centralizes visibility into who used AI, through which system and how much.
All the systems request work through the same gateway. Adding AI and moving cloud work use the same entry point. Beyond it, the unit’s LLM completes processing internally.
See usage in one place
The AI-OS provides user controls and audit logs recording actions. Work entering through the gateway follows shared controls, avoiding separate platform-management implementations for every system.
Add capacity with another unit
In supported multi-unit configurations, another unit on the same network can be discovered and added to the cluster. Connected systems keep using the same gateway, centralizing expansion. Confirm configuration requirements and capacity with sales.
One reception desk centralizes visitor records, passes and security. A separate back door for every department would multiply the recording and protection work. A single API gateway provides one reception desk for AI.
Where should you start? Three steps, beginning with one system
You do not have to connect everything at once. Start with one frequent, repetitive task. When the result is measurable, expand to another system.
List current AI calls and work that could benefit from AI
Use cloud AI bills to see which systems call AI and how much. Then list the summaries, drafts and comparisons staff perform manually every day. These two lists identify candidate integrations.
Choose frequent, repetitive work with internal data
Typical examples include inquiry summaries and questions about internal policies. After required initial setup, the unit can be used for a trial. If a compatible integration only needs an endpoint change, major redevelopment may not be necessary. Validate the actual scope and output quality.
Measure the effect and move to the next system
Record changes in cloud bills and work time. The shared gateway can make later integrations easier. If concurrency grows, assess additional capacity.
| Current situation | Use case to start with | First example |
|---|---|---|
| Cloud AI bills rise each month | Move suitable cloud workloads internally | Redirect a frequently used compatible integration |
| Business systems do not yet use AI | Add AI to an existing system | Add summarization to inquiry handling |
| Confidential data cannot go to cloud AI | Add AI while keeping data internal | RAG over internal policies or design documents |
| Both needs exist | Move suitable cloud workloads and expand AI-enabled workflows | Use savings to fund additional integration work |
Cloud AI bills rise each month
- Use case to start with
- Move suitable cloud workloads internally
- First example
- Redirect a frequently used compatible integration
Business systems do not yet use AI
- Use case to start with
- Add AI to an existing system
- First example
- Add summarization to inquiry handling
Confidential data cannot go to cloud AI
- Use case to start with
- Add AI while keeping data internal
- First example
- RAG over internal policies or design documents
Both needs exist
- Use case to start with
- Move suitable cloud workloads and expand AI-enabled workflows
- First example
- Use savings to fund additional integration work
Illustrative starting points; the best choice depends on your organization.
Common misconceptions and the actual arrangement
API integration can sound like a major development project. The distinction is that you use an existing AI capability rather than develop the AI itself. Actual integration effort still depends on the system.
| Misconception | Actual arrangement |
|---|---|
| Adding AI requires rebuilding the existing system | Integrate a call to the gateway; the whole system need not be rebuilt |
| Switching from cloud AI requires changing every request format | Compatible cloud-style calls can target the gateway; validate supported features and integration behavior |
| API usage incurs a separate per-call charge | Japanese one-time price: JPY 3,806,400 before tax, without extra generation charges; operating expenses and options are separate |
| System integration sends internal data elsewhere | The gateway can route to the local LLM, completing processing internally |
| It only helps if all cloud AI is abandoned | A mixed approach works; moving frequent routine tasks can change cost growth |
| Every new connected system creates separate platform administration | One gateway centralizes user controls and audit records |
Actual arrangement
- Adding AI requires rebuilding the existing system
- Integrate a call to the gateway; the whole system need not be rebuilt
- Switching from cloud AI requires changing every request format
- Compatible cloud-style calls can target the gateway; validate supported features and integration behavior
- API usage incurs a separate per-call charge
- Japanese one-time price: JPY 3,806,400 before tax, without extra generation charges; operating expenses and options are separate
- System integration sends internal data elsewhere
- The gateway can route to the local LLM, completing processing internally
- It only helps if all cloud AI is abandoned
- A mixed approach works; moving frequent routine tasks can change cost growth
- Every new connected system creates separate platform administration
- One gateway centralizes user controls and audit records
Common API-gateway questions based on published information as of September 2026
Mini glossary: terms used in this article
API gateway (Sovereign GaiXer)
A shared entry point through which internal systems and apps request work from the unit’s AI.
API
A defined interface through which systems request work from each other. Like an order counter, it accepts a request in an agreed format and returns a result.
Generative AI / LLM (large language model)
In this article, AI that reads and produces text. The model behind it is an LLM, which Sovereign GaiXer hosts inside the unit.
Token
A unit used to count text processed by AI. The Japanese source uses an illustrative estimate of about 1.5 tokens per Japanese character; actual counts vary by model and text.
Metered billing / token cost explosion
Charging according to token usage, and the rapid bill growth that can occur when systems call AI automatically at scale.
Cloud AI
AI running on another organization’s servers accessed over the internet, often with usage-based charges.
Internal AI (on-premises / local AI)
AI running on a unit within the company, with processing completed internally. Sovereign GaiXer follows this approach.
Adding AI to existing systems
Adding generative AI capabilities to a system that previously had none. With Sovereign GaiXer, the system integrates with the API gateway.
Offloading
Moving work previously handled by cloud AI to an internal unit, by redirecting compatible calls to the API gateway.
LLM gateway / hybrid AI
An LLM gateway receives and routes AI requests through one entry point. Hybrid AI combines cloud and internal AI according to the task. The single-gateway approach in this article follows these principles.
RAG (internal-data retrieval)
Answering while referring to internal documents ingested into the unit, without sending that data outside the company.
API gateway: frequently asked questions
What is Sovereign GaiXer’s API gateway?
It is a shared entry point through which internal systems and apps ask the unit’s AI to perform work. Instead of a person using a screen, a system requests a summary or answer in the background and receives the result. The same gateway can add AI to systems that had none or move existing cloud AI workloads to the local unit.
Can we add generative AI without rebuilding an existing system?
Yes. The system itself need not be rebuilt, but it needs an integration change to call the gateway. You use the AI already in the unit instead of developing AI from scratch. Discuss the scope of the change with your system owner or developer.
Can a system already using cloud AI switch to it?
The gateway supports cloud-style API calls, allowing compatible integrations to change their endpoint to Sovereign GaiXer. The requester then communicates with an internal unit instead of an external server. Confirm the specific API features, model behavior and integration requirements for your application before switching.
Does API use add charges?
Under the Japanese offer, Sovereign GaiXer is a one-time purchase of JPY 3,806,400 before tax, with no additional charges for more generations or users, excluding electricity and other operating expenses. API call volume does not create metered usage bills. Confirm overseas pricing and terms with sales.
What is a token cost explosion?
It describes a rapid increase in metered cloud AI bills. Human chat limits request frequency, but integrated systems may call AI whenever an inquiry or order arrives. Call volume can become many times greater, increasing charges with token usage. The Japanese source uses a rough illustrative estimate of 1.5 tokens per Japanese character; actual tokenization varies by model and text.
Do we need to stop using cloud AI entirely?
No. A practical combination uses cloud AI for advanced analysis of public information and Sovereign GaiXer for confidential, high-volume, repetitive work. Moving frequent routine workloads internally can change how costs grow without abandoning cloud AI.
Does internal data leave when systems use the API?
The gateway routes local inference to the LLM inside the unit, where processing is completed. Answers can be generated without transmitting inputs or referenced internal data to the internet. See disconnected operation on the security page.
Does connecting several systems make administration harder?
A shared gateway centralizes user management and audit records in the AI-OS, so each connected system does not need to rebuild those platform capabilities. Integration-specific controls still need to be configured appropriately.
How many systems can one unit connect to?
There is no specified limit on the number of connected systems, but more simultaneous generations can increase latency. Multi-unit configurations can add capacity: units on the same network discover each other and cooperate without changing each connected system’s endpoint. Configuration and capacity requirements should be discussed with sales.
Where should we begin?
Start with one system. Review cloud AI bills and the summaries or drafts staff write manually each day, then choose one frequent, repetitive task. Record results in both billing and work time before expanding to the next system. Reusing the gateway can make subsequent integrations easier.
Three points to understand the shared API gateway
- Add generative AI to existing systems. Integrate calls to the gateway without rebuilding the whole system. Use the AI already in the unit rather than build it yourself.
- Move suitable metered cloud AI workloads to internal processing. Redirect compatible integrations and validate them. The Japanese one-time purchase price is JPY 3,806,400 before tax, with no additional per-call usage charge, excluding operating expenses. Confirm overseas conditions with sales.
- Both use the same entry point. One gateway centralizes user management, audit logs and capacity expansion.
Put one AI reception desk inside your company: that is Sovereign GaiXer’s API gateway. Systems from any department can bring requests to it. Suitable work previously sent outside on a pay-per-use basis can go through the same internal desk. One reception point means one place for shared records and controls, even as more systems connect.
Related reading
What is Sovereign GaiXer? The AI-OS through a smartphone analogy
Understand PGX, AI-OS, app store and business apps, and the capabilities of the AI-OS.
AI and Money seriesTokens, metered billing and the metered-billing trap
Why cloud AI costs can be hard to predict: a plain-language explanation for decision-makers.
Product guide, Part 5What if your network blocks internet access?
Enable updates and web search on a restricted LAN, compare three approaches and prepare a change request.
This article reflects the Japanese source as of September 2026. Confirm API connection methods, supported features and concurrency guidance with sales before implementation.Integration scope varies by your system architecture. Comparisons describe general metered cloud AI scenarios, not specific products or recommendations for or against particular providers. “About 1.5 tokens per Japanese character” is an illustrative estimate that varies by text and tokenizer. “No additional charge” refers to the unit’s usage model, excluding electricity and other operating expenses; paid options such as GaiXer Care are separate. The monthly equivalent allocates a Japanese pre-tax purchase over three years and is not a monthly subscription. All listed prices describe the Japanese offer; overseas pricing and commercial conditions require confirmation. “AI-OS” describes the software foundation for secure business AI. Smartphone, cafeteria and reception analogies explain product structure without claiming compatibility or equivalence with particular products. ThinkStation is a Lenovo trademark. Other company, product and service names are trademarks or registered trademarks of their owners.