Models, ready from the factory
BixinAI selects today's advanced local models compatible with S1, preinstalled at the factory so model inference runs on the device.
LOCAL INFERENCE NIUCLAW S SERIES / Local compute agent host
NiuClaw S1NiuClaw S1
Let AI compute power,
be right beside you.
Local inference, local storage; local AI can run even when disconnected.
Offline use, data never leaves the device, privacy fully in your control.
Factory-preinstalled curated local models and NiuClaw agent or well-known agents such as WorkBuddy.
Local inference requires no Token purchase and is not billed by Token.
Windows by default, so AI is ready out of the box.
Market Price¥32,999yuan
DESIGNED FOR YOUR AI WORKSPACE
LOCAL AI COMPUTINGLocal compute · Desktop deployment
360° PRODUCT VIEW
From another angle,
Meet S1.
Drag the chassis to view different angles.
You can also switch to actual product photos at any time.
01 / LOCAL INTELLIGENCE
The next step for agents,
is to have compute by your side.
Continuing the office agent experience of the NiuClaw series,
S1 further provides local compute, supporting the host in
running local model inference.
Work, in familiar Windows
Runs Windows by default, placing everyday office software, files, and agents in a familiar work environment.
WINDOWS WORKSPACEAgents, ready out of the box
Ships with NiuClaw agent software. From software configuration to getting started, the official guide walks you through the first step.
View the NiuClaw user guide ↗HARDWARE / Core hardware configuration
The confidence behind local AI,
comes from solid hardware.
CPU, GPU, and NPU compute in coordination,
paired with large memory and flexible expansion,
providing the foundation for running local models.
AMD RYZEN™ AI395 configuration
Ryzen AI Max+ 395
AMD Ryzen AI processor
- Combined compute
- 126 TOPS
- LPDDR5X memory · up to
- 128 GB
- VRAM capacity*
- 96 GB
- Storage expansion
- Confirm according to selected configuration
- Wired network
- 2.5G + 10G dual Ethernet ports
- USB connection
- 6 × USB-A + 2 × Type-C
- Display output
- 1 × HDMI + 1 × DP
- Chassis dimensions
- Compact desktop chassis
- System and software
- Windows by default
Capacity and expansion specifications are subject to the confirmed configuration sheet.
2 × RJ45 · subject to markings on the device
Both front and rear have Type-C; rear is labeled USB4
See the rear product image for display port layout.
Specific dimensions are subject to the new model's configuration sheet.
Factory-preinstalled NiuClaw agent software
Hardware specifications are listed according to the 395 configuration in product materials; actual supplied configuration is subject to the selected option. Memory is the maximum supported capacity; the number of storage bays does not represent the number or capacity of factory-installed drives. * VRAM capacity follows the labeling in the materials; actual usable capacity is subject to the actual configuration and settings; TOPS is a combined compute metric and does not represent the generation speed of any specific model.
02 / ON-DEVICE COMPUTING
Worry-free local Tokens,
No need to pay for every inference.
Local models included out of the box; no need to purchase Tokens.
Local inference is not billed by Token, with no Token recharge quota limits.
Factory-selectedBixinAI continuously evaluates currently advanced models and selects models that balance intelligent capabilities and S1 hardware compatibility for factory preinstallation.
Continuous updatesOfficial continuous model update maintenance keeps up with more advanced, compatible models; users can also update or replace local models themselves.
01Input taskFrom your local apps
→02S1 local inferenceCall deployed model
→03Return resultContinue your workflow
The above Token-free benefit refers to running preinstalled local models on this device; actual running speed and capacity are affected by hardware and model configuration. If cloud models or third-party paid services are selected, their fees are calculated according to the corresponding service rules.
YOUR LOCAL MODEL
NIUCLAWS1LOCAL INFERENCE
Compute · Models · ApplicationsYOUR LOCAL AI WORKSPACE
Let local AI,
become part of everyday desktop life.
NiuClaw S1 · dedicated compute,
Connecting work, exploration, and creation.
03 / AGENT ECOSYSTEM
A familiar system.
Open agent choices.
Windows by default, pre-installed with NiuClaw.
It also supports the office agents you are used to using.
N↗Factory pre-installed
YOUR BUILT-IN AGENT
NiuClaw agent
Once the host arrives, the agent software is already in place.
Learn how to use and configure the software through the official user guide.
WINDOWS office agent ecosystemInstall and configure as needed
WorkBuddy
Supports WorkBuddy office agent
Qwen Office
Supports Qwen Office agent
Doubao Work
Supports office agents such as Doubao Work
Third-party software must be installed, logged in to, and configured according to its requirements; functions and services are subject to the descriptions provided by each software provider.
LOCAL AI WORKSPACE / Industry and creative scenarios
Data has boundaries,
and AI should too.
Case, R&D, business, and personal data,
deserve a work environment you control.
When an organization requires sensitive business data to stay on designated devices, intranets, or project environments, S1's local compute power and pre-installed models provide the foundation for a dedicated AI workstation. From industry data management to local video creation, choose your scenario and see which specific task you can start with.
NIUCLAW S1 / 01
One repository per case makes experience retrieval more organized.
Scenario illustration · solution needs adaptationSolo practitioners · boutique law firms · litigation teams
Case experience stays in your own case file repository.
Materials for entrusted cases should not be uploaded casually to public AI; you want to reuse experience from past years while avoiding cross-client mixing.
- Data you need to protect
- Engagement contracts, evidence materials, client communication records, lawyer case notes
- AI work that can be planned
- Organize case timelines; compile evidence lists and items to verify; search authorized historical case notes.
Example task · not an actual demonstration result
“Please extract a timeline from the materials in this case, cite the source document for each item, and list facts still requiring manual confirmation.”
Case repositories and access permissions must be configured; facts, citations, and legal judgments are reviewed by lawyers.
NIUCLAW S1 / 02
Spend less time digging through files and more energy on service and processing.
Scenario illustration · solution needs adaptationNon-classified government office work · public institution offices · government service departments
Internal materials, organized within an approved environment.
Public information and internal business data have clear usage boundaries; without approval, they cannot be handed to external services for processing.
- Data you need to protect
- Non-classified internal documents, meeting minutes, processing ledgers, policy materials
- AI work that can be planned
- Search policies within approved scope; organize meeting items; generate processing material checklists and preliminary official document drafts.
Example task · not an actual demonstration result
“Generate a processing material checklist based on local policy materials, cite the source of clauses, and flag conflicts in materials for review.”
Only non-classified scenarios are discussed. Unit approval and adaptation assessment are required; no claim is made that this product has classified qualifications or government compliance certification.
NIUCLAW S1 / 03
Let important materials form reusable project assets.
Scenario illustration · solution needs adaptationCorporate legal · board office · M&A project team
Major projects do not need to be uploaded to the cloud first.
Trade secrets and undisclosed decision-making materials need to be confined to environments approved by the project team and organization.
- Data you need to protect
- M&A due diligence documents, undisclosed business plans, contract versions, board materials
- AI work that can be planned
- Compare contract versions; summarize due diligence issues; organize decision material summaries and to-do lists.
Example task · not an actual demonstration result
“Compare these two versions of the procurement contract, list changes in payment, liability, and termination clauses, and point to their source locations.”
Authorization is per project; output only assists organization, and business and legal conclusions still require review by the responsible person.
NIUCLAW S1 / 04
Newcomers can find information quickly, and experience is retained.
Scenario illustration · solution needs adaptationR&D engineers · quality team · process department
Process know-how becomes an internal knowledge asset.
Core technical materials have commercial value; uploading them to external tools often requires special authorization first.
- Data you need to protect
- Undisclosed design specifications, process parameters, test reports, defect analyses, supplier specifications
- AI work that can be planned
- Search test records; summarize troubleshooting experience; compare specification documents and quality rectification records.
Example task · not an actual demonstration result
“Organize the causes, verification steps, and handling results of the last three similar defects, with corresponding test records attached.”
Drawings, CAD, and complex tables require separate adaptation and parsing; AI does not replace engineering verification or safety decisions.
NIUCLAW S1 / 05
Create images and videos locally, and keep sensitive materials in-house.
Scenario illustration · Image and video models requiring deployment adaptationCross-border sellers · DTC website teams · Brands expanding overseas
Create product images and marketing videos locally,
while keeping business data under your control.
New product launches require product images, campaigns require videos, and continuously testing multiple versions consumes generation credits; customer information, suppliers’ lowest prices, and unreleased products are not suitable for casual uploading to external AI. Let NiuClaw S1 handle both local content production and business analysis.
- Image production: product displays and marketing materials
- Deploy and adapt open-source image models to create lifestyle images, advertising creatives, and marketing visuals from product images, repeatedly experimenting with styles and backgrounds for use on product detail pages and in campaigns.
- Video production: save on cloud generation API costs
- NiuClaw S1 locally deploys and adapts open-source video generation models to turn product images or text concepts into product showcase clips and short advertising video materials. Local generation requires no per-use cloud video API fees; create multiple shots and versions repeatedly while reducing reliance on paid generation credits.
- Data security: process sensitive materials locally
- Customer and order information, supplier quotations, cost and profit reports, and unreleased product materials are stored and processed locally; once the models and workflows have been deployed, imported information and materials can be processed offline.
- Business efficiency: analyze reports and organize after-sales support
- Consolidate business reports from multiple stores, verify procurement quotations, and analyze product gross margins; summarize after-sales issues and draft multilingual customer service replies.
Example task · not an actual demonstration result
“Use this new product image to create three lifestyle scenes, then generate product showcase videos with different shots; process the materials locally only, for the team to review, edit, and use in campaigns.”
Image and video models and workflows require separate deployment and adaptation; local generation still incurs equipment, electricity, and storage costs, while speed and results depend on the model and configuration. Commercial use must comply with the model license. Store synchronization and campaign management require an internet connection; data permissions, product authenticity, and external-facing content require human confirmation.
NIUCLAW S1 / 06
Materials are more organized, and review is more focused.
Scenario illustration · solution needs adaptationAccounting firms · tax service teams · corporate internal audit
Client working papers, organized securely by project.
Client business information and practice materials must be kept confidential, and cross-access between projects must be controlled.
- Data you need to protect
- Audit working papers, financial statements, contract vouchers, tax materials, internal control records
- AI work that can be planned
- Summarize working paper key points; check against the voucher checklist; generate materials to be supplemented and exception review items.
Example task · not an actual demonstration result
“Based on this project’s voucher catalog, list missing items and numbers to be checked; do not fill in missing amounts on your own.”
Numerical verification, audit judgment, and report issuance are performed by professionals.
NIUCLAW S1 / 07
Sort out the materials first, then let professionals make the decisions.
Scenario illustration · solution needs adaptationApproved business pilot · Risk control support · Compliance and internal control team
Customer data, used within the authorized scope.
Customer financial information and business data are subject to strict access, entrusted processing, and audit requirements.
- Data you need to protect
- Customer application materials, credit attachments, claims materials, internal control standards
- AI work that can be planned
- Within the approved pilot, verify material completeness; search internal control policies; summarize issues pending manual review.
Example task · not an actual demonstration result
“Mark missing documents against the local application materials checklist; do not infer customer credit ratings or make automated approval decisions.”
Subject to institutional approval and security assessment; does not replace business systems or approval decisions for credit, claims, and similar operations.
NIUCLAW S1 / 08
Leave the organizing work to the tool, and leave judgment to medical staff.
Scenario illustration · solution needs adaptationApproved in-hospital pilot · Medical record management · Medical quality control support
Medical record organization, with patient privacy protected first.
Health information is sensitive; use of materials requires a legal basis, authorization, and in-hospital management procedures.
- Data you need to protect
- Medical record text, discharge records, quality control materials, hospital policies
- AI work that can be planned
- Organize the medical record text index; extract document items to be checked; search hospital standards and authorized quality control materials.
Example task · not an actual demonstration result
“Extract missing document items from authorized medical record text and indicate the source; do not generate diagnoses or treatment recommendations.”
Limited to document and knowledge organization assistance; does not make diagnosis, medication, or treatment decisions; requires institutional approval and professional review.
NIUCLAW S1 / 09
Everyday Q&A is easier, with less personal information circulating.
Scenario illustration · solution needs adaptationCorporate HR · Payroll team · HR shared services
Employee data, only in the hands of those who need it.
Employee information has purpose and access restrictions and must not be freely provided to unrelated persons or tools.
- Data you need to protect
- Payroll sheets, labor contracts, performance materials, employee grievances, and personnel files
- AI work that can be planned
- Search internal HR policies; organize a list of expiring contracts; summarize employee inquiry materials within the authorized scope.
Example task · not an actual demonstration result
“Answer annual leave process questions based on local policies, cite policy clauses, and do not retrieve other employees’ salary information.”
Tiered authorization for payroll and files; do not use AI output directly to make hiring, performance, or termination decisions.
NIUCLAW S1 / 10
Try ideas repeatedly, without paying for every generation.
Scenario illustration · Requires deploying a compatible video modelSelf-media creators · E-commerce merchants · Brand marketing teams
Open-source video models, create freely locally.
Want to try a few more shots and make a few more versions, but always have to count generation counts and credit consumption? Deploy a compatible open-source video model to S1 and let ideas be tried repeatedly.
- No per-generation payment, unlimited creations
- Run free open-source video models locally, with no need to buy cloud Tokens or generation credits and no cloud platform generation quota limits.
- Videos you can create
- Generate product display clips, short-video storyboards, and brand creative assets from text or images; repeatedly adjust visuals and shots, then edit into a final video.
- Assets stay local
- After deploying the model and dependencies, generate offline using local assets; unreleased product images and creative plans need not be uploaded to the cloud.
Example task · not an actual demonstration result
“Use this product image to generate a display video with a slow push-in shot, trying warm-light and cool-light atmospheres respectively for editing selection.”
Requires separately deploying video models and workflows compatible with S1; no per-generation payment refers to local inference, with no cloud generation fees, but there are still device, electricity, and storage costs. Generation speed, duration, and resolution depend on the model and configuration; commercial use must comply with the chosen model's license.
KEEP THE WORKFLOW LOCAL
Need “not leave the domain,”
then keep the entire workflow within the domain.
Store data locally → Parse and search locally → Local model inference → Authorized personnel use results
Document parsing, OCR, vector retrieval, plug-ins, logs, and backups also need to be localized according to the plan; disable external model and networked tool calls, configure access permissions, network egress controls, and an approved model update method, then accept according to the actual environment.The above are application planning examples, not delivered cases or industry systems available out of the box. Confidentiality obligations do not mean all materials are legally prohibited from leaving the domain; boundaries are determined by data classification, authorization, contracts, and organizational policies. Local inference does not automatically equal zero external transmission or compliance certification, and S1 is not to be used as a classified system; government, financial, medical, and other scenarios must first pass institutional approval and adaptation assessment.
Scenario basis and applicability notes
Confidentiality and security requirements refer to the following public regulations; specific application scenarios are solution projections based on these requirements and do not represent certification or endorsement of the product by the relevant institutions.
Lawyer's duty of confidentiality ↗Government Data Sharing Regulations ↗Trade Secret Protection ↗Code of Ethics for Certified Public Accountants ↗Measures for Data Security Management of Banking and Insurance Institutions ↗Measures for Cybersecurity Management of Medical and Health Institutions ↗Personal Information Protection Law ↗
DEDICATED DEVICE / The value of a dedicated device
Advantages of NiuClaw compared with software assistants such as WorkBuddy
Also let AI do the work,
What difference does a dedicated device bring?
From local computing power to home learning,
Eight scenarios, clear at a glance.
NiuClaw S1 Local independent computing powerNiuClaw N1 / N1 mini Memory stays local
Below compares two approaches: “assistant running on a daily office laptop” and “tasks independently deployed on NiuClaw.” S1 also supports office agents such as WorkBuddy; cloud tasks are not directly affected by laptop sleep. The image is a scenario illustration.
01
Local models need their own computing power
Software assistant on an office laptop
The software itself does not provide independent computing power
Installing an assistant on an office computer still requires computer hardware or cloud services to provide model inference computing power.
NiuClaw dedicated device
S1 comes with its own computing power and its own local models
Preinstalled with selected local models at the factory; on-device inference requires no Token purchase and is not billed by Token.
S1 local computing power
This item corresponds to S1. Cloud models and third-party paid services are billed separately according to their rules.
02
The laptop enters sleep after the lid is closed
Software assistant on an office laptop
Tasks that depend on the local machine are paused
Local tasks running on an office laptop will pause when the computer sleeps.
NiuClaw dedicated device
Close the laptop lid, and your dedicated device keeps working
Tasks are configured independently on NiuClaw and are not interrupted when the office laptop goes to sleep.
Dedicated device
The device must remain powered, connected to the network, with sleep disabled and services running, and tasks must be configured independently.
03
WeChat, Tencent Meeting, and CapCut open at the same time
Software assistant on an office laptop
AI and office software share resources
When the agent also opens web pages and processes files, lag is more likely when memory is tight.
NiuClaw dedicated device
S1 takes on tasks, reducing the load on the office computer
Dedicated compute handles AI tasks deployed on S1, while your computer focuses on meetings and editing.
S1 local computing power
This item corresponds to S1; the corresponding tasks need to be deployed to S1 for execution.
04
Long-term memory and personal data management
Software assistant on an office laptop
AI environment and daily office work share the same device
Installed on the office laptop, the AI's local environment and daily use remain on the same device.
NiuClaw dedicated device
Local memory, stored on a dedicated device
N1 / N1 mini stores agent memory locally, accumulating long-term context on a dedicated device.
N1 / N1 mini local memory
Local memory is not the same as local model inference; how models are called depends on the actual configuration.
05
Want to set up a dedicated working environment for AI
Software assistant on an office laptop
A host and runtime environment still need to be arranged
If you want the software assistant to run independently and always on, you need to prepare a separate device and configure the runtime environment.
NiuClaw dedicated device
Hardware and software bundled together, reducing setup effort
Default Windows, with NiuClaw pre-installed at the factory; together with skills and task management, it builds a dedicated working environment.
Hardware and software bundled
Specific tasks, accounts, and tools still need to be configured according to actual use scenarios.
06
Take the laptop out to meet clients
Software assistant on an office laptop
Local tasks still depend on the office laptop
In connected-computer mode, local tasks require this office computer to stay online and running.
NiuClaw dedicated device
You take the computer with you, while AI keeps working at home
Through the configured WeChat, Feishu, DingTalk, or remote desktop entry point, continue to assign tasks and receive results.
Remote use
An available remote entry point must be configured in advance, and the home device and network must remain operational.
07
Separate personal AI exploration from workplace office work
Software assistant on an office laptop
Personal exploration still occupies work equipment
Personal interest research and daily creation share the same computer and runtime environment with workplace office work.
NiuClaw dedicated device
Personal exploration, on your own device
NiuClaw stays at home, the workplace computer focuses on work, and personal data and creations have their own space.
Dedicated space
Separating devices to organize personal and office environments does not mean network or data isolation is automatically achieved.
08
The whole family learns and uses AI together
Software assistant on an office laptop
Sharing an office computer requires taking turns
Adults working and children learning use the same computer, so device time needs coordination.
NiuClaw dedicated device
A dedicated family device makes learning easier to arrange
Parents accompany children to practice asking questions, coding, and creating, with learning materials and works organized in one place.
Family learning
Usage time and access methods are arranged according to family needs; children's use is recommended to be accompanied by parents.
04 / FIND YOUR NIUCLAW
The same agent starting point,
with an extra measure of local compute.
The core difference between S1 and N1, N1 mini,
is the compute capability for local model inference.
| Product capabilities | The choice for local computeNiuClaw S1 | NiuClaw N1 | NiuClaw N1 mini |
|---|---|---|---|
| Product positioning | Local compute agent host | Office agent host | Office agent host |
| Local model inference | Supported · Local compute | Not positioned for local inference | Not positioned for local inference |
| Default operating system | Windows | Windows | Windows |
| Pre-installed software | NiuClaw agent | NiuClaw agent | NiuClaw agent |
| Other office agents | WorkBuddy / Qwen Office / Doubao Work, etc. | WorkBuddy / Qwen Office / Doubao Work, etc. | WorkBuddy / Qwen Office / Doubao Work, etc. |
NIUCLAW S1 / NiuClaw S1
From every side,
meet your AI work partner.
Silver body, black front panel,
bringing dedicated compute into your everyday desktop.
NIUCLAW S1 / PRODUCT DETAIL05 / YOUR AI DESK
On the desktop,
one more AI work partner.
Daily office work and local AI exploration unfold on the same Windows host.
- 01
Office agent workstation
Use NiuClaw or other supported office agents to build your own way of working around daily tasks.
- 02
Local model exploration
Deploy compatible local models and try AI applications such as text generation and document processing.
- 03
AI application development and validation
Provide a desktop runtime environment for applications that require local inference, and validate the compatibility of models and workflows.
06 / GOOD TO KNOW
You might want to know.
What is the main difference between S1 and N1, N1 mini?
S1 provides local inference compute power and comes preinstalled with compatible local models selected by BixinAI; local inference requires no Token purchase. All three products run Windows by default and come with NiuClaw agent software preinstalled.
What software comes preinstalled on S1?
It comes with Windows by default, preinstalled with NiuClaw agent software and compatible local models selected by BixinAI. Other office agents such as WorkBuddy, Qwen Office, and Doubao Work are also supported and can be installed and configured as needed.
Where can I view the instructions for the NiuClaw agent?
Visit NiuClaw agent official user guide ↗, to learn how to use and configure the software.
Can all office agents use local models directly?
S1's local compute power provides the hardware foundation for local model deployment. Whether specific software can connect to local models depends on the model interfaces and configuration methods it supports; support for installing office agents does not mean that all functions of the software can run offline.
Do local models still require purchasing Tokens?
No. Using the local models preinstalled on S1 for on-device inference requires no Token purchase, is not billed by Token, and has no Token top-up limit. Actual operating speed, context length, and the scale of tasks that can be handled depend on the hardware and model configuration; if you additionally use cloud models or third-party paid services, charges apply according to their respective rules.
Will the preinstalled models be updated? Can I replace them myself?
BixinAI continuously evaluates currently advanced models, selects models that balance intelligence capabilities and compatibility with S1 hardware to preinstall at the factory, and continuously maintains model updates. Users can also update or replace compatible local models themselves as needed.
How do I choose a configuration and a suitable local model?
You can consult us based on the model scale, task types, and operating requirements you plan to use, to confirm the compatible hardware configuration and deployment plan.
NIUCLAW S1 / READY FOR LOCAL AI