1. User & Input Layer
| English | 中文 |
|---|---|
| User Request: The user asks the AI to perform a task such as writing, analyzing, or generating content. | 用户请求:用户要求 AI 执行任务,例如写作、分析或生成内容,这是整个 AI 流程的起点。 |
| Prompt: The detailed instruction given to the AI describing what to do and any constraints. | 提示词:用户给 AI 的具体指令,说明要做什么以及限制条件,用来精确控制输出。 |
| Prompt System: Defines the AI’s role, tone, and behavior rules. | Prompt 系统:定义 AI 的角色、语气和行为规则,例如“你是数据工程师,要用专业语气回答”。 |
2. AI Core Reasoning Layer(AI 核心推理层)
| English | 中文 |
|---|---|
| LLM Reasoning: The model interprets the request, understands intent, and performs logical reasoning. | LLM 推理能力:模型理解用户意图,并进行语义解析与逻辑推理,是 AI 的核心“思考能力”。 |
| AI Brain: The central decision-making unit that determines how to solve the task step by step. | AI 大脑层:系统级决策核心,负责判断任务怎么做、是否需要查资料、是否调用工具。 |
| Planning Module: Breaks complex tasks into structured steps for execution. | 规划模块:将复杂任务拆解成多个步骤,例如“先查数据 → 再分析 → 再总结”。 |
3. Knowledge & Information Layer(知识与信息层)
| English | 中文 |
|---|---|
| Context: Background information such as conversation history and input data used to understand the request. | 上下文信息:包括聊天历史和背景数据,用来帮助 AI 理解当前问题的来龙去脉。 |
| Retrieval (RAG): Retrieves relevant information from enterprise knowledge bases before generating answers. | 检索增强生成(RAG):在生成答案前,从企业知识库或文档中检索相关信息,避免“凭空编造”。 |
| Enterprise Knowledge: Internal company data such as documents, databases, and policies used as trusted sources. | 企业知识库:企业内部数据来源,例如文档、数据库、制度等,是 AI 回答的真实依据。 |
| Memory & State: Tracks conversation history, task progress, and user context over time. | 记忆与状态:记录对话历史和任务进度,让 AI 能“记住之前发生的事情”。 |
4. Tools & Execution Layer(工具与执行层)
| English | 中文 |
|---|---|
| Tool Calling: Allows the AI to interact with external systems such as APIs, databases, or enterprise tools. | 工具调用:让 AI 能调用外部系统(API、数据库、企业工具)来完成真实操作。 |
| Enterprise Actions: Enables AI to perform real-world actions such as writing data, sending emails, or triggering workflows. | 企业动作能力:AI 可以执行真实动作,例如写入数据、发送邮件或触发系统流程。 |
| Enterprise Workflow: Structured business processes where multiple steps are automated and orchestrated by AI. | 企业流程编排:将多个业务步骤串联起来的自动化流程,例如审批流或数据处理流程。 |
| Enterprise Operations: System-level operations including monitoring, scheduling, and resource management. | 企业运营系统:负责系统运行的底层能力,包括监控、调度和资源管理。 |
5. Safety, Quality & Governance Layer(安全与治理层)
| English | 中文 |
|---|---|
| Guardrails: Rules that enforce safety, compliance, and prevent harmful or invalid outputs. | 安全护栏:控制 AI 行为边界,防止违规内容、敏感信息泄露或错误操作。 |
| Evaluation: Measures output quality, correctness, and detects hallucinations. | 质量评估:检查 AI 输出是否正确,是否存在“幻觉问题”,以及是否符合标准。 |
| Observability: Monitors system performance, latency, errors, and overall AI pipeline health. | 可观测性:监控整个 AI 系统运行状态,包括延迟、错误率和性能指标。 |
Overall System Summary(整体总结)
| English | 中文 |
|---|---|
| User Request triggers the system, Prompt System defines behavior, LLM performs reasoning, Planning breaks tasks into steps, RAG retrieves knowledge, Tools execute actions, and Governance ensures safety and quality. | 用户请求触发系统,Prompt 系统定义行为规则,LLM 负责推理,Planning 负责拆解任务,RAG 提供知识,工具执行动作,最后由安全与治理层保证系统的安全与质量。 |

