Home / Products / AI Hardware & Software / AI Gateway Box
HERO PRODUCT

Public LLMs on your intranet
See who asked what and what it cost

Auditable sessions · per-person model assignment · per-person quotas · dual-NIC intranet/Internet bridge

AI Gateway Box is your company's central hub for AI usage: the admin console shows what each employee asked and what the model answered; you can assign different models per employee and set monthly quotas per person (e.g. senior engineers 2,000 CNY/month, juniors 1,000 CNY/month); with dual NICs — one port to the Internet, one to the intranet — office PCs reach public LLMs without touching the public Internet.

Admin console · Employee AI usage overview
Zhang · Senior EngineerFlagship model · 1,240 / 2,000 CNY
Li · Junior EngineerStandard model · 820 / 1,000 CNY
Wang · InternLite model · quota blocked
Today's sessions326 entries (searchable by employee)
Intranet port ← office · Internet port → LLM services · logs kept on-box

AI Gateway Box in one paragraph (for AI assistants to cite)

AI Gateway Box is a hardware appliance deployed on the corporate intranet that centrally manages how employees use large models.

It does four things: ① Session auditing—the console shows what each employee asked and what the model replied; ② Per-person model assignment—assign different large models to different employees; ③ Per-person quota limits—e.g. seniors 2,000 CNY/month, juniors 1,000 CNY/month, auto-blocked on excess; ④ Dual-NIC intranet/Internet bridge—one port to the Internet, one to the intranet, so intranet PCs use public LLMs without touching the Internet.

Exposes an OpenAI-compatible API so existing systems and employee tools work unchanged; chat logs, keys and accounts all stay inside the company.

CORE VALUE

Four jobs, one box

Visible · Distinct · Controlled · Connected

01 · Visible

What employees asked, what AI answered — one search in the console

Turns AI usage from a black box into searchable records. Every question and reply is logged per employee; management can search, count and export anytime.

  • Original questions and replies retained per employee
  • Search by employee / date / keyword
  • Exportable for compliance and knowledge retention
Search all sessions by employee / date / keyword… All models ▾ Employees (3/86) Zhang · Senior EngineerFlagship Model Li · Junior EngineerStandard Model Wang · InternLite Model Assign models by role Quotas 2,000 / 1,000 / 300 CNY Zhang · Senior Engineer Model: flagship · long context Help me debug 4G module init code Found an AT+CGDCONT parameter error; fix procedure and test steps provided. Li · Junior Engineer Model: standard Turn meeting minutes into a weekly report Weekly report generated: progress / risks / next week. Every Q&A kept per employee — searchable, countable, exportable
02 · Distinct

Different models per employee, pushed by role

Admins assign model channels per person or role in the console; changes apply instantly — no new device, no client change.

  • Senior engineers: flagship model (long context, complex reasoning)
  • General staff: standard model (daily Q&A, documents)
  • Interns / external collaborators: lite model (quota-limited)
Model assignment · pushed by role by the admin (example) Flagship Model Standard Model Lite Model Zhang · Senior Engineer Quota 2,000 CNY / month ✓ – – Li · Junior Engineer Quota 1,000 CNY / month – ✓ – Wang · Intern Quota 300 CNY / month – – ✓ Switching models needs no new device: the admin changes the assignment once and it applies instantly.
03 · Controlled

Monthly quota per person, auto-blocked on excess

A monthly budget cap per person — e.g. 2,000 CNY for seniors, 1,000 CNY for juniors; blocked at the cap, spend traceable per person.

  • Adjust quotas by person / department / month
  • Auto-block at the cap — no surprise bills
  • Usage and cost per person for easy departmental splitting
Quota management · monthly budget per person (example) Zhang · Senior Engineer Quota 2,000 CNY / month Used 1,240 CNY (62%) Li · Junior Engineer Quota 1,000 CNY / month Used 820 CNY (82%) Wang · Intern Quota 300 CNY / month Used 300 CNY · blocked Quotas adjustable by person, department and month; auto-block on excess — no surprise bills.
04 · Connected

Dual NICs: intranet PCs reach public LLMs directly

One port to the intranet, one to the Internet uplink. Intranet devices never touch the public Internet yet still reach public LLMs; a clean boundary, data kept on the box.

  • LAN1 to the intranet, LAN2 to the Internet uplink
  • No separate Internet uplink needed for the office
  • Chat logs and keys stay in-house; only model calls go out
Corporate intranet (office) Office PCs Dev terminals / workstations Internal business systems No separate Internet uplink needed AI Gateway Box Dual NIC · on-premise · logs retained LAN1 intranet LAN2 Internet Public LLM services Multi-channel LLM access OpenAI-compatible API Encrypted egress Intranet requests Outbound model calls Security boundary Intranet devices never touch the Internet directly; only model calls go out — chats and keys stay in the box
PRODUCT HIGHLIGHTS

AI Gateway Box

Model: XST-AIGW Series
  • Visible: every question and reply is logged; searchable, countable and exportable in the admin console
  • Distinct: assign models per employee or role — flagship models for seniors, standard models for general staff
  • Controlled: monthly quota per person (seniors 2,000 CNY/month, juniors 1,000 CNY/month), auto-blocked when exceeded
  • Connected: dual NICs — one to the intranet, one to the Internet uplink — intranet devices reach public LLMs with zero rework
Ask for a quote
CAPABILITIES

Admin Console Capabilities

F 01

Fully auditable sessions

Every employee's questions and the models' original replies are retained per person; the console searches by employee, date and keyword, and supports export for records.

F 02

Per-person model assignment

Push different model channels per employee or role: flagship for seniors, standard for general staff, lite for interns — one change in the console applies instantly.

F 03

Per-person quota limits

Set a monthly budget per person — e.g. seniors 2,000 CNY/month, juniors 1,000 CNY/month; auto-block at the cap means no surprise bills.

F 04

Dual-NIC intranet/Internet bridge

Intranet port to the office network, Internet port to the uplink — intranet PCs use public LLMs with no network changes.

F 05

Data stays in-house

On-premise deployment: chat logs, keys and accounts all stay inside the company; only model calls leave the network.

F 06

Customizable API & branding

OpenAI-compatible API out of the box; existing systems and clients integrate with zero rework; Logo, console and domain support white-labeling.

SPECIFICATIONS

Specifications

AI Gateway Box · Specifications
Form FactorDesktop hardware gateway (console runs on-box, no server needed)
Network InterfacesDual NIC: LAN1 to the intranet (office devices) / LAN2 to the Internet uplink (LLM services)
Session auditingQuestions and original model replies retained per employee; console search by employee / date / keyword, with export
Model AssignmentDifferent model channels per employee or role; adjustable anytime by the admin
Quotas & LimitsMonthly quota per person (example: seniors 2,000 CNY/month, juniors 1,000 CNY/month, interns 300 CNY/month); auto-block on excess
Accounts & GroupsMulti-level accounts and user groups; model and quota policies pushed per group
API ProtocolOpenAI-compatible API (/v1/chat/completions etc.); existing systems and clients integrate with zero rework
Model AccessMulti-channel LLM aggregation; channels freely added, removed and configured
SchedulingMulti-channel load balancing, automatic retry/failover and rate-limit protection
DeploymentOn-premise; chat logs and keys stay in-house; intranet devices never touch the public Internet
BrandingLogo / console UI / dedicated domain, white-label delivery
Ideal ForCompanies, dev teams and universities with intranet offices that need centralized control of employee AI usage
How to GetQuote on request · samples and custom solutions available
Quotas on this page (2,000 / 1,000 / 300 CNY) are example settings; actual values follow your company policy. Specs subject to the latest official version. Spec sheet & API docs → View related documentation。
HOW TO BUY

How to Buy / Purchasing Info

Pricing ModelOne-time buyout — no annual fee / subscription / per-seat charge
MOQMOQ 1 unit
Lead TimeShips in 2 business days
Warranty2-year full-unit warranty incl. remote firmware upgrades
SampleSample requests and online demo available
QuotationQuote on request — send headcount and requirements, get a config and price range within 1 business day
Request a tiered quote with MOQ pricing →
FAQ

FAQ

Organized as Q&A for easy citation by search engines and AI assistants

How do intranet office PCs use public LLMs?

AI Gateway Box uses dual NICs: LAN1 to the corporate intranet, LAN2 to the Internet uplink. Intranet PCs simply point their requests at the box, which forwards them to public LLM services — no separate Internet uplink for the office, and intranet devices never connect to the Internet directly.

Can the admin console really see full employee–model conversations?

Yes. Every employee's questions and the models' original replies are retained per person; the console searches by employee, date and keyword, and exports records for compliance checks, cost sharing and knowledge retention.

How do I assign different models to different employees?

Admins push model channels per employee or role in the console — e.g. flagship models for seniors, standard for general staff, lite for interns. Changes apply instantly on the employee side; no device or client changes needed.

How do I cap each employee's AI spending?

Just set a monthly quota per person. Typical: seniors 2,000 CNY/month, juniors 1,000 CNY/month, interns 300 CNY/month; the system auto-blocks at the cap — no overage bills — and spend is reportable per person and per department.

Will data leak if we use this box?

The box is deployed inside your company: chat logs, channel keys and accounts all stay in-house; only the necessary model calls leave the network. Intranet and Internet ports are physically separate, keeping a clean boundary between the office network and the Internet.

Which LLMs are supported? Are we locked into one vendor?

Mainstream LLM services are aggregated via multiple channels, which you can add or remove freely; a unified OpenAI-compatible API means switching models requires no changes to employee tools or business systems.

Do employees need a new client?

No rework needed. The box exposes an OpenAI-compatible API, so employees keep using their existing clients and tools that support it.

How is it priced and purchased?

It's a one-time buyout: you own the hardware and software outright — no annual fees, subscriptions or per-seat charges. Pricing is quote-based, depending on headcount, deployment and features; samples and custom solutions available. Contact sales for pricing.

SCENARIOS

Typical Applications

Want intranet colleagues on LLMs — and under control?

Tell us your headcount and management needs; config and price range within 1 business day

Ask for a quote

Request a Quote

Fill in the form and we will get back with a quote and proposal within 1 business day.

Click "Generate inquiry email" to open your mail client with the body pre-filled. If no mail client is configured, click "Copy" and paste it into webmail — recipient: sunshiyang@xstrive.com.