As large language models and agent technologies continue to advance, enterprises are accelerating their exploration of AI applications in knowledge management, office collaboration, and business processes — driving internal knowledge reuse and business efficiency gains.
Around real business needs, enterprises expect to quickly validate the value of AI applications at lower investment cost and with more convenient deployment, and then gradually scale proven capabilities to more business scenarios.
However, most AI implementation solutions on the market lean toward large-scale, full-featured platform construction — with numerous components, complex deployment, and demanding requirements for hardware resources and operations staffing.
Enterprises that only want to validate a few business scenarios end up having to invest significant resources in building the environment and completing integration — bearing substantial construction costs before seeing any application results.
To lower this implementation barrier, ZStack and Alibaba Cloud have jointly launched the ZStack + OpenTrek lightweight AI implementation solution.
The olution combines Alibaba Cloud OpenTrek's knowledge and agent application capabilities with ZStack Zaku's container runtime, rsource management, and supporting services. Through integrated collaboration between the application layer and the underlying foundation, it simplifies component integration, deployment, and operations — enabling customers to start lightweight and scale on deand, moving AI from business validation toward sustained application.
This is another joint solution release by the two parties, following their cloud-edge collaboration solution.
As a strategically controlled subsidiary of Alibaba Cloud, ZStack leverages Zaku in this collaboration to handle the runtime and management needs of AI applications, complementing Alibaba Cloud's upper-layer AI capabilities to provide customers with support spanning from AI application deployment to continuous operation.
01 / Lightweight AI Deployment
From application to runtime foundation — a lightweight, on-demand, composable AI implementation solution
A knowledge assistant or business agent requires knowledge processing, model invocation, workflows, runtime environment, and data services to work together.
The ZStack + OpenTrek lightweight AI implementation solution organizes product capabilities around these dependencies, allowing customers to select components by business need and obtain the corresponding runtime support.
At the application layer, OpenTrek provides Model-as-a-Service (MaaS) and agent capabilities, covering model access, knowledge base, data processing, agent development, application publishing, and operations monitoring.
ZStack Zaku, as a lightweight AI Infra, provides enterprise-grade Kubernetes clusters and RDS middleware services. Through cluster management, resource scheduling, elastic scaling, networking, storage orchestration, and monitoring, it supports the unified deployment and continuous operation of agent services and related components.
The underlying resource environment can be flexibly selected based on the customer's existing infrastructure:
- When physical servers are available, Zaku can be deployed directly on them to build a Kubernetes cluster;
- When virtualization resources need to be consolidated, Zaku can be deployed on virtual machines provided by ZStack Cloud;
- Customers with an existing third-party virtual machine environment can also deploy Zaku on the basis of meeting compatibility and resource requirements, using the existing environment to host AI applications.
When GPUs are involved, the specific support for passthrough, transparent transmission, or vGPU depends on the compatibility between the hardware and the virtualization platform.
On this architecture, OpenTrek connects models, knowledge, tools, and application management — providing complete capability support for enterprises to build private AI digital employees.
02 / OpenTrek AI Capabilities
Alibaba Cloud OpenTrek: connecting model services, agent development, and application operations
Alibaba Cloud OpenTrek provides a privately deployable Model-as-a-Service (MaaS) and agent platform, covering model access, fine-tuning adaptation, knowledge processing, agent building, application publishing, and operations monitoring.
Enterprises can select components around their business needs, combining model capabilities, internal knowledge, and business tools into deliverable, manageable AI applications.
1. Model access and services — supporting the capabilities required for different task invocations
Alibaba Cloud OpenTrek supports access to large language, multimodal understanding, vector, and reranking model services, providing support for content generation, knowledge retrieval, and agent reasoning.
Alibaba Cloud OpenTrek-AI Gateway provides a unified invocation entry, combining model routing, rate limiting, token metering, and content safety capabilities to help enterprises manage model usage across different applications.
Through unified model access and invocation management, application teams can select models around their tasks, and administrators can stay informed of usage — providing a basis for subsequent application expansion and resource investment management.
2. Knowledge processing and retrieval — turning enterprise materials into AI-usable knowledge
Alibaba Cloud OpenTrek provides document libraries, data tables, image-text libraries, and online and offline parsing capabilities to process and organize enterprise materials scattered across different carriers, and to provide business grounding for models and agents through knowledge retrieval.
Knowledge capabilities can be combined with Q&A entry points, workflows, and tool invocations to support applications such as document lookup, content organization, and business assistance.
Enterprises can bring long-accumulated policies, product documentation, technical materials, and service experience into AI applications. Supported by material updates and permission management, they can continuously expand the scope of knowledge use and improve internal knowledge reuse efficiency.
3. Agent development and execution — organizing models, knowledge, and tools into business applications
Alibaba Cloud OpenTrek provides workflow orchestration and autonomous planning agent development capabilities. Combined with prompt engineering, tool invocation, memory management, and skill management, it supports designing task execution processes according to business needs.
Fixed steps can be organized through workflows, while tasks requiring flexible handling can be planned through agents and access business systems via already-integrated tools.
Alibaba Cloud OpenTrek-Agent Runtime provides a runtime environment and sandbox capabilities to carry agent and tool execution
Development teams can reuse validated processes and skills for similar tasks, progressively accumulating application capabilities suited to different roles and reducing repetitive development and configuration work.
4. Application publishing and operations — supporting agents in continuous production use
Alibaba Cloud OpenTrek incorporates application publishing, hosting, monitoring, and analysis into its platform capabilities. Through logs, call-chain tracing, and other means of observing application runtime, it helps teams locate issues in model invocation, tool execution, and task processing.
Combined with account, permission, and gateway controls, enterprises can manage application access and the scope of capability use, and continuously adjust configurations based on runtime feedback.
From development to launch, enterprises can reuse the corresponding publishing and management mechanisms, moving agents from one-off validation toward everyday use. Specific capabilities can be deployed in combination according to the selected modules and versions.
03 / Zaku Cloud-Native Foundation
ZStack Zaku: connecting application deployment, resource scheduling, and continuous operations
Dovetailing with OpenTrek's model services and agent capabilities, ZStack Zaku provides an enterprise-grade Kubernetes cluster and a cloud-native runtime environment. It brings container orchestration, compute management, data service support, and operations capabilities into a unified foundation — supporting AI applications through the full journey from deployment and launch to scaling and upgrade.
1. From foundation to supporting components — upgrades and refinements for lightweight deployment
To help enterprises truly achieve lightweight AI implementation, ZStack has made targeted optimizations to the foundation, supporting components, and delivery stages around OpenTrek's deployment needs.
① Streamlining the foundation to lighten deployment preparation
The installation package has been slimmed from approximately 20GB to approximately 5GB. Combined with a lightweight Kubernetes foundation, this reduces the burden of installation media transfer and distribution. For the large images common in AI applications, upload speeds have been optimized to make deployment preparation and application delivery more convenient.
② Refining lightweight supporting services to reduce piecemeal integration
A lightweight object storage is provided to support OpenTrek's persistent data needs, along with database and middleware services such as MySQL, PostgreSQL, Redis, and Elasticsearch.
At the same time, compatibility has been implemented with Xinchuang databases such as PolarDB, OceanBase, and Dameng (DM), providing database selection and deployment support for customers with domestic-innovation construction requirements.
Customers can select supporting components according to application needs, organizing deployment and updates through Zaku's image management, Helm application management, and container orchestration — reducing the work of self-building and integrating dependent components.
③ Turning adaptation work into delivery support
By combining product version adaptation, standardized deployment, OpenTrek delivery capability import, and delivery training, the solution connects the implementation process between the foundation and upper-layer applications — reducing repetitive configuration and communication coordination in projects, and extending lightweight implementation from product deployment to project delivery.
2. Resource scheduling and management — supporting on-demand allocation and scaling
Zaku provides resource scheduling, tenant quotas, GPU management, and runtime status observation, supporting resource allocation by application demand, insight into GPU health and utilization, and resource expansion as business grows.
Combined with validated adaptation combinations of domestic operating systems, data components, and computing power, customers can flexibly select based on their existing environment and construction requirements, matching application scale to resource investment.
3. Unified operations and services — supporting long-term application operation
Zaku provides a view of cluster and resource operation through monitoring, logging, alerting, and inspection, helping operations teams detect anomalies and troubleshoot issues. Supporting capabilities such as data protection, permission management, and upgrades provide a foundation for routine maintenance and platform evolution.
Combined with standardized deployment, joint delivery services from both parties, and original-manufacturer support, customers receive continuous service handoff from environment preparation and joint-debugging launch through to follow-up maintenance — reducing the management burden of multi-component, multi-team collaboration.
As AI applications increase, enterprises can reuse Zaku's deployment, resource, and operations systems to extend their existing environment, reducing the investment of repeated setup and integration when adding new scenarios.
04 / AI in Business Practice
Into customer business scenarios — putting AI capabilities to work
Today, the ZStack + OpenTrek lightweight AI implementation solution has been validated in multiple real customer scenarios.
Through the collaboration of OpenTrek and Zaku, the solution connects upper-layer AI applications with underlying runtime management, providing support for customers moving from business validation toward sustained application.
Around needs such as knowledge query and business collaboration, the joint solution can form a rich variety of application patterns, exemplified by the following two scenarios:
Scenario 1: Enterprise knowledge assistant.
By bringing policies, product manuals, or technical documentation into the knowledge base, employees can query procedures, product information, or operating methods through natural language.
Alibaba Cloud OpenTrek handles material processing, retrieval, and answer generation — with the citation capability of selected components supporting source verification. Zaku, together with database and storage support, carries the related services and supports material updates and application expansion.
Enterprises can start from high-frequency query needs, reducing repeated searching and redundant consultation, and making internal knowledge more conveniently available for everyday work.
Scenario 2: Business collaboration assistant.
Around tasks such as project progress reports, Alibaba Cloud OpenTrek can orchestrate material retrieval, business system queries, and content organization into a workflow: reading authorized materials, obtaining progress through integrated tools, generating a draft report in the agreed structure, and then handing it to business staff for verification and refinement.
Zaku provides application deployment, resource allocation, and runtime monitoring support, enabling validated processes to be used continuously and extended to more teams as business requires — reducing the work of repeated collection and organization.
05 / Business Value and Growth
Let business outcomes guide investment; let AI capabilities accumulate continuously
To advance AI, enterprises need to judge in actual use which tasks are worth investing in and which applications are worth scaling.
Lowering deployment and integration burdens allows business teams to participate in validation earlier, using whether knowledge queries are more convenient, whether repetitive work is reduced, and whether processing results meet requirements as the basis for the next round of investment — letting the pace of construction follow business value.
As applications deepen, organized knowledge, validated workflows, and integrated tools can be reused within applicable permissions and business scope; the runtime foundation can also continue to carry new applications.
The ZStack + OpenTrek lightweight AI implementation solution lets enterprises explore AI at more controllable cost, while keeping the valuable results of each validation round within their own business system — reducing the investment of building subsequent applications from scratch.
Want to make enterprise knowledge available on demand, and free employees from repeatedly looking up materials, compiling information, and organizing reports? Looking for a private AI application solution that can quickly validate business results while controlling upfront investment?
Welcome to contact ZStack to discuss your business needs with us, and explore together how AI can be implemented in real-world work.