Article body
Full article
HENGSHI Data Agent can understand intent, break down complex data questions, and support metric creation, dashboard creation, and intelligent interpretation within authorized data scope. ChatBI is better suited to an independent intelligent Q&A entry point and to generating charts and dashboards in a conversation. The two can be composed through product entries, APIs, or an enterprise’s own workflows, but a fixed built-in handoff object, unified state, and automatic delivery are not default behavior in every deployment. This article treats the path as an implementable solution pattern and separates public capabilities from project-integration work.
1. Why coordinate two engines?
ChatBI suits standalone intelligent Q&A and can continuously create charts and dashboards around a conversation. It should not be reduced to a component that can only answer simple questions; its task range depends on connected data, model configuration, and current product capabilities.
Data Agent can determine intent from user input, break down complex questions, and retrieve and query assets within the user’s authorized data scope. When an enterprise needs to orchestrate results into approvals, report delivery, cross-system actions, or recurring operating tasks, its own workflow, API integration, or HENGSHI CLI orchestration and scheduling capabilities must define task boundaries, authorization, and human review.
The value of coordination is to connect natural-language interaction, governed analysis, and auditable execution. HENGSHI provides ChatBI, Data Agent, APIs, bot integration, and HENGSHI CLI as capability surfaces. Enterprises can design a path from understanding and querying to creation, approval, and delivery. The extent of automation and the checkpoints that require confirmation should follow business risk and project permission design.
2. Optional implementation stages
2.1 Frame the request and scope
For a composite analytics request, define the analysis object, time range, metric definition, data sources, delivery form, and approver. Data Agent can understand input and decompose complex questions, but whether ChatBI asks follow-up questions automatically and which entry collects recipients and delivery channels are front-end and workflow-integration choices, not one uniform default interaction.
2.2 Define a task contract for external orchestration
When a result is handed to an external Agent, workflow, or backend system, the integrator should define a clear task contract: objective, data scope, definitions, input constraints, deliverable, recipient, and timing. HENGSHI provides Data Agent APIs for backend orchestration and cross-domain orchestration capabilities in HENGSHI CLI. Public materials do not specify a fixed built-in object named Task Spec between ChatBI and Data Agent.
Such a contract can be an integration agreement between ChatBI, Data Agent, external workflows, and backend systems. It is project-defined rather than a fixed HENGSHI internal object. Select the handoff method according to system boundaries, API capabilities, the permission model, and maintainability.
2.3 Analyze and execute within authorization
Data Agent can run one or more data queries according to question complexity and retrieve assets within authorization. For data connections, metric modeling, dashboard creation, permission actions, alerting, and scheduling, HENGSHI CLI provides commands and Skills that Agents can call. The toolchain, validation steps, and whether an asset can be written must be governed by project permissions and human review.
If intermediate progress must be shown, teams can use a streaming Data Agent HTTP API or return task state to the front end from an enterprise workflow. Progress presentation differs across a ChatBI page, Data Agent sidebar, API, and bot entry, so validate each surface before implementation.
2.4 Present results and retain human review
Results can appear in a ChatBI conversation, dashboard, rich-text report, or an integrator’s own page. HENGSHI supports chart and dashboard creation in ChatBI conversations, while Data Agent can assist with visualization and metric creation. Review of definitions, data sources, and conclusions should be part of the delivery process; no fixed confidence value or report shape should be promised for every result.
2.5 Deliver and track when authorized
Recurring analytics, mobile notifications, and external delivery are actions that need explicit authorization. HENGSHI data integration and CLI include scheduling and orchestration surfaces, and public HENGSHI BOX cases show resident Agents monitoring metrics and sending analysis briefings to DingTalk and WeCom. Email, system messages, approvals, and recipient permissions still require confirmation against the enterprise’s existing systems and integration design.
3. The technical foundation for collaboration
When analysis and execution are coordinated, ChatBI, Data Agent, and external workflows should reference consistent metrics, data scope, and definitions. HENGSHI metric management and NL2Metrics provide a basis for semantic governance; whether entries share the same data package or context must be designed and verified explicitly.
ChatBI conversation context, Data Agent page context, API calls, and external-workflow state do not necessarily share a boundary. To carry a task between entries, define context identifiers, state storage, permission revalidation, and failure recovery in the integration design. CLI, API, bot, and external-workflow identities also need their own least-privilege scope, data scope, and recipient checks rather than assuming they inherit the front-end user’s permissions.
4. Typical closed-loop scenarios
For anomaly attribution, a governed flow can include dimensions such as periods, channels, and campaigns; Data Agent can run one or more queries, then business users review the charts and conclusions. For periodic reviews, existing scheduling, HENGSHI CLI, or external workflows can produce sales overviews and review materials on a schedule. For decision support such as store feasibility, connected sales, footfall, and location data can be combined with an enterprise’s own scoring model and sensitivity analysis. Each scenario needs distinct design for data sources, business rules, conclusions, delivery permissions, and human review.
5. Design principles
Divide responsibilities by entry point and task boundary rather than treating ChatBI and Data Agent as mutually exclusive. ChatBI is suited to standalone questions and conversational chart or dashboard creation. Data Agent is suited to analysis, visualization, and metric creation in page context and can decompose complex questions. Cross-system automation should be orchestrated explicitly through APIs, CLI, or workflows.
Introduce human review or approval at risk points such as objective confirmation, asset writes, consequential conclusions, and external delivery. Start with governed Q&A, metric creation, or dashboard creation, then extend to alerts, scheduling, and cross-system workflows as evidence accumulates. Automation level, execution identity, and delivery channel all need individual configuration and acceptance.
| Capability area | Public HENGSHI foundation | What requires project integration or acceptance |
|---|---|---|
| Standalone Q&A | ChatBI creates charts and dashboards around a conversation | Data scope, model configuration, embedded entry |
| In-page analysis and creation | Data Agent can analyze, visualize, and assist metric creation from page context | Page, version, and authorization scope |
| Complex questions | Data Agent can decompose questions and run one or more queries | Business rules, review of conclusions, cross-system actions |
| Engineering execution | CLI covers data, modeling, dashboards, permissions, scheduling, and orchestration | Skill configuration, write permissions, Dry Run, human review |
| Delivery and sustained operation | Automation can be designed with APIs, bots, and scheduling | Recipients, approvals, state sync, notification, retention |
6. From one goal to a controlled outcome
ChatBI, Data Agent, APIs, bots, and HENGSHI CLI are not isolated products that must be wired into one fixed architecture. They are capability surfaces that can be combined by scenario. Semantic governance, permission boundaries, and auditable execution are the common foundation. State handoff, approval, cross-system actions, and delivery channels must be made explicit in the integration design. This is how intelligent analytics can grow from a demonstration into controlled business infrastructure.