Huawei Cloud Zhiguo is a full-stack intelligent agent development platform launched by Huawei Cloud, positioned as the "last mile" engineering foundation connecting large models with industry scenarios. It originates from Huawei's experience in orchestrating complex systems in telecommunications and government-enterprise services, integrating Pangu large model's multimodal reasoning capabilities to provide enterprises with a progressive development environment ranging from low-code visual drag-and-drop to high-code Python/API deep customization. Unlike simple large model API aggregation tools, Zhiguo's core is building an observable, rollback-able, and auditable agent runtime engine that supports multi-agent collaboration, dynamic tool invocation, and knowledge base enhanced generation (RAG), with Huawei Cloud's full-chain security protection system built in.

In terms of industry standing, Zhiguo has passed the CAICT intelligent agent platform special evaluation, deployed over 200 production-grade cases in vertical fields such as finance, manufacturing, and government affairs. Its "model-neutral" strategy allows simultaneous access to third-party models like Pangu, GPT, and Llama, forming a differentiated neutral ecosystem position. The platform emphasizes governance-first, providing full-chain lineage tracking from intent recognition to execution decisions, and has been rated by third-party institutions as a core product in the leader quadrant of China's intelligent agent development platforms.

• Visual workflow orchestration - The drag-and-drop canvas supports complex flows such as sequential, branching, parallel, and loops, with 30+ built-in nodes (code execution, API calls, human approval, condition judgment), allowing business personnel to build cross-system automation without programming.
• Multi-model gateway and hybrid routing - Unified access to mainstream models such as Pangu, OpenAI, and Claude, with intelligent routing strategies based on task difficulty and cost thresholds, automatically balancing response quality and inference overhead.
• Enterprise-grade knowledge base enhancement (RAG) - Seamlessly connects to Huawei Cloud OBS, KooSearch, and mainstream vector databases, providing document parsing, vectorization, hybrid retrieval, and citation tracing, supporting 20+ formats including PDF/Word/tables, ensuring answer explainability.
• Human-machine collaboration with manual review - For high-risk operations (fund transfers, data deletion), mandatory manual review nodes can be set, supporting Webhook integration into enterprise approval flows to meet financial compliance requirements.
• Observability and full-chain tracing - Built-in Agent Trace feature visualizes each node's input/output, token consumption, latency, and failure chains, supporting log replay and error retry to accelerate troubleshooting.
• High-code custom extension - Provides Python SDK and plugin mechanisms, allowing developers to write custom tools and custom model adapters, embedding complex algorithm logic into flows to achieve unlimited capability boundaries.

• Step 1: Log in and create a workspace - Visit the official website and log in with your Huawei Cloud account. After entering the console, select the "Zhiguo" service, click "Create Workspace," and choose a personal workspace or an enterprise team workspace (team workspaces allow asset sharing and member role permission configuration).
• Step 2: Start from a template or blank canvas - The platform includes 20+ business templates such as "Customer Service," "Data Analysis," and "Content Production" that can be directly copied and modified; or start from a blank canvas, drag nodes on the left, and configure node parameters on the right (select model, fill in prompts, bind databases, etc.).
• Step 3: Debug and test - Click "Test Run" in the upper right corner, input test data, and observe intermediate node outputs; use "Debug Mode" to execute step-by-step and locate error nodes; adjust prompts or flows and save as a new version.
• Step 4: Publish and integrate - After testing, click "Publish" to generate an API Endpoint or Webhook address; you can set scheduled triggers or embed it into your own application via SDK; monitor call volume, success rate, and cost consumption on the monitoring dashboard.

Example 1: Intelligent ticket classification and auto-reply (customer service scenario)
On the canvas, connect the "Ticket System - Event Trigger" node → call the Pangu NLP model with a prompt template: "Based on the ticket content, output a JSON format category (technical fault/billing issue/product inquiry/complaint) and urgency level (high/medium/low), and generate a polite initial reply script." → Add a conditional branch: if urgency is high, automatically notify the duty manager (via WeCom bot). Expected output: tickets are classified within 2 seconds, high-priority tickets trigger reminders, and reply drafts are generated for manual confirmation.

Example 2: Automated sales weekly report generation (marketing scenario)
Use "Scheduled Trigger" to execute every Friday at 6 PM: call API to pull this week's opportunity data from CRM (e.g., Salesforce) → use data analysis node to calculate conversion rate and won amount → then call the large model to generate an HTML weekly report with a prompt template: "Based on the following data [{data}], output a sales weekly report containing core metric trends, highlight cases, and risk warnings, with professional and concise language." → Finally, send it to the sales director via email node. Expected output: a well-structured report with complete chart notes, saving 90% of manual compilation time.

Example 3: Contract clause risk review (legal scenario)
Upload a PDF contract to the "Document Parsing" node → extract text and call the large model with a prompt template: "You are a senior legal counsel. Review each clause of the following contract, identify liability clauses unfavorable to Party A, vague monetary expressions, missing breach clauses, output risk levels and provide modification suggestions in a table format." → Enhance with historical precedents from the knowledge base and generate a risk review report. Expected output: automatically flag 3 high-risk clauses and 5 medium-risk points, with legal basis attached.

• Full-chain domestic autonomy and controllability - Deeply optimized based on Huawei Ascend computing power and Pangu models, supporting full domestic deployment from chips to platforms, meeting the requirements of party and government, finance, and other industries for autonomous and controllable IT systems.
• Flexible model-neutral strategy - Not bound to a single model, allowing mixed calls to models from different vendors to avoid vendor lock-in; also provides a unified abstraction layer, so switching models requires zero changes to business logic.
• Enterprise-grade security and compliance governance - Supports private deployment, VPC isolation, data encryption, and desensitization, with operation auditing and full-chain logs. It has passed Level 4 cybersecurity classification protection and SOC 2 certifications, making it more suitable for highly regulated industries compared to international competitors.
• Seamless collaboration between low-code and high-code - Business personnel can complete 80% of workflows through drag-and-drop, while developers supplement complex logic with Python/API, and the two can be converted to each other, unlike most platforms that favor only one side.
• Deep integration with Huawei Cloud ecosystem - Out-of-the-box connectivity to 200+ Huawei Cloud services (databases, function computing, OCR, speech synthesis), seamless integration with ModelArts and CodeArts, facilitating the construction of AI-native applications.

Comparison with mainstream platforms like Coze, Dify, and Alibaba Cloud Bailian:

• Compared with Coze (ByteDance) - Zhiguo emphasizes enterprise governance and private deployment, supporting complex manual approval flows and audit trails, while Coze leans toward C-end consumer plugin markets; Zhiguo is more mature in multi-agent collaboration and hybrid routing.
• Compared with Dify - Dify is an open-source-friendly tool platform; Zhiguo is stronger in visual orchestration capabilities, high-concurrency performance, and cloud-native elastic scaling, and comes with Huawei Cloud security compliance certifications; for non-developers, Zhiguo's prebuilt templates have finer industry granularity.
• Compared with Alibaba Cloud Bailian - Both provide large model application development on the cloud, but Zhiguo's model-neutral strategy is more thorough (Bailian focuses on Tongyi), and Zhiguo has significant inference cost optimization in the Ascend ecosystem, suitable for cost-sensitive large-scale call enterprises.

Zhiguo adopts a hybrid commercial model of "free basic version + pay-as-you-go + enterprise subscription":

• Free Tier - Provides 500 free agent runs per month (limited time), including canvas orchestration, basic model calls (Pangu trial), and community support, suitable for personal learning and prototype validation.
• Pay-as-you-go - After exceeding the free quota, charges are based on agent invocation count, runtime duration (CPU/memory), and model token consumption. Pangu models have additional discounts per token, with instance runtime billed per second (starting at approximately 0.001 CNY/second).
• Pro / Team versions - Offer monthly packages. Pro is approximately 199 CNY/month (includes 10,000 runs + 100GB private knowledge base), Team is approximately 999 CNY/month (supports 5 collaborative members, shared assets, permission management, and API SLA of 99.95%).
• Enterprise version (customized) - Supports private deployment, dedicated computing resources, and dedicated technical support, with annual contracts. Pricing requires contacting sales, based on node count and concurrency evaluation.

• Enterprise digital transformation leaders (business architects) - Need to automate business flows but are not professional programmers; they use low-code orchestration to quickly build intelligent agents for approval, customer service, reporting, etc., to achieve cost reduction and efficiency improvement.
• AI application development engineers and algorithm engineers - Use high-code SDKs and custom model interfaces to embed model capabilities into core business systems and use observability to troubleshoot performance bottlenecks.
• IT operations and security administrators - Focus on private deployment, permission auditing, and data compliance; Zhiguo's support for domestic IT systems and full-chain tracing helps meet regulatory requirements.
• Industry solution consultants - Quickly build intelligent agent prototypes for POC demonstrations in finance, government, and manufacturing projects to accelerate customer signing.

• Intelligent customer service and ticket automation - Automatically handle common inquiries (e.g., billing queries, fault reports), dynamically access enterprise knowledge bases, and transfer complex issues to humans with context summaries.
• Data analysis and report generation - Regularly extract data from databases/CRM, call analysis nodes and large models to generate visualized weekly reports and anomaly alert emails, with support for natural language query metrics.
• Content production and multilingual localization - Combine crawlers/uploaded materials to automatically generate marketing copies and social media posts, and call translation nodes to produce multilingual versions while maintaining brand tone consistency.
• Operations and IT process orchestration - Integrate monitoring systems, automatically trigger fault diagnosis scripts, call large models to analyze logs and generate troubleshooting suggestions, automatically execute allowed operations (e.g., service restart), or submit tickets for manual approval otherwise.
• Education and legal intelligent Q&A - Provide intelligent Q&A assistants based on private documents for universities or enterprise training, with source citations to ensure answers are verifiable.