This topic is for readers in the solution evaluation phase. It describes the differentiated capabilities of DataPilot compared to similar products, item by item. Each entry includes a capability description, implementation method, and practical benefits for users, along with the corresponding documentation entry.
Natural Language Data Retrieval
Capability: Business users can ask questions to the Agent in natural language, and the Agent generates queries based on the semantic layer and authorized data scope.
Implementation Method: The semantic layer organizes authorized data tables into business objects, attributes, relationships, and metrics. In a single request, the Agent combines models, authorized data scopes, semantic assets, and tools to translate questions into queries constrained by data policies.
Benefit: Business users can retrieve data independently without developers writing SQL or building reports in advance, and the scope of questions is controlled by the boundaries of authorized data.
Related Documentation: Overview of Semantic Layer, Start Dialog Analysis
Conversational Generation for All Workspaces
Capability: Ten types of workspaces—reports, dashboards, big screens, generative data applications, multi-dimensional tables, Notebooks, workflows, spatio-temporal analysis, document classifiers, and data extraction factories—can all be generated through Agent-driven conversation. After generation, they can be further refined via conversational iteration or workspace editing according to their type.
Implementation Method: In a single request, the Agent combines models, authorized data scopes, semantic assets, and tools to write results into the corresponding application objects. Data applications are released and authorized alongside the Agent.
Benefit: Business personnel can obtain editable application objects upon requirement submission, without waiting for development schedules. Engineers maintain all deliverables within the same set of data boundaries.
Related Documentation: Conversational Data Applications, Start Dialog Analysis
Agent Marketplace and Subscription Governance
Capability: After release, an Agent becomes a discoverable asset. Analysts can subscribe directly or submit an application, which is approved by administrators based on workspace policies. When an enterprise integrates with an external permission system, trusted integration services can also directly convert completed authorization conclusions into subscriptions.
Implementation Method: Release status, visibility scope, subscription records, and specific resource permissions are four independent judgments, managed by Agent engineers, release configuration, workspace administrators, and resource owners respectively.
Benefit: Analytical capabilities can flow within the organization like assets while retaining hierarchical authorization checkpoints. Successful subscriptions do not bypass data policies.
Related Documentation: Agent Marketplace and Subscription, Subscription Approval
Dual-Role Workspaces
Capability: Engineers enter the build workspace to create and maintain objects from the Agent list. Analysts enter the marketplace and My Agent to use only published versions.
Implementation Method: Roles determine the default workspace and visible entry points, while permissions determine specific actions. The user menu provides an entry to switch roles, and the current context changes accordingly after switching.
Benefit: The two types of users do not interfere with each other or see entries unrelated to them. Engineers' drafts are not accidentally exposed to users.
Related Documentation: Roles and Reading Paths
Native Multi-Dimensional Tables and AI Columns
Capability: Multi-dimensional tables organize data from data sources or files into worksheets that can be filtered, sorted, and grouped. AI columns call models based on the input of each row to generate text, classification, or structured fields, and can depend on other regular columns, derived columns, or AI columns.
Implementation method: Dependencies determine the execution order; cache strategies reuse existing results that meet conditions; after confirming results, current computed results can be solidified and written to the target data source to form stable replicas.
Benefits: Turning "model processing" into a column in a table allows for small-scale trial runs to control costs, and once results stabilize, they can be solidified for downstream stable use.
Related documents: AI columns, Solidified data views, and Analyst personal columns
Ontology Semantic Layer
Capability: Semantic processing organizes authorized data tables into business objects, attributes, relationships, dimension value aliases, metrics, and executable functions, providing confidence levels and evidence within the ontology topology.
Implementation method: Synchronization is a long-running background task, supporting resumption from breakpoints and full rerun; metric calibers are jointly defined by computation logic and filter conditions; JSONL test cases can be used to batch evaluate the "question-to-metric" matching, obtaining pass/fail results and accuracy rates for each item.
Benefits: Business calibers shift from personal experience to reviewable, testable, and testable assets, serving as the foundation for question-answering accuracy.
Related documents: Overview of the semantic layer, Semantic layer construction checklist
Multi-layer Data Permissions
Capability: Data policies restrict access to tables, rows, and columns based on members or member attributes. Agents and data applications only read authorized data.
Implementation method: Policies are maintained at the workspace level, supporting both local policy and external data permission hosting modes; regular preview returns bounded samples, permission preview applies policies according to the selected preview identity, and external hosting mode follows the hosting configuration; adding a data table to an Agent's data scope does not bypass policies.
Benefits: The same Agent can automatically present different data scopes for different positions without requiring a duplicate configuration for each position.
Related documents: Data permission model, Permission design practice
End-to-end Workflow
Capability: Data processing, model invocation, and message notification are orchestrated in a node-based manner. After passing testing, a new version takes effect via a released Agent, initiating scheduling.
Implementation method: Workflows save drafts; after passing testing, a new version takes effect via a released Agent; published workflows can be triggered by authorized users, running facts are imported into the workflow task domain of the task center.
Benefits: Repetitive analysis and distribution actions are solidified into schedulable processes, allowing for unified observation of runtime status and failure phases.
Related documents: Overview of workflows, Version and enablement management
Sandboxing Execution and Hierarchical Observability
Capability: Notebook and code nodes execute in a controlled sandbox; runtime observability provides four complementary types of facts: Monitor summarizes health and sessions, Audit records who performed what operations, system logs provide service-side output, and the task center summarizes the lifecycles of long-running tasks such as semantics, knowledge indexes, AI columns, workflows, and batch tests.
Implementation method: Monitor supports observation scopes at both workspace and global levels; system logs are organized by service source and region, and error traces can be exported.
Benefits: When troubleshooting, it is possible to distinguish between "symptoms" and "causes", avoiding confusion between configuration changes, service errors, and task failures.
Related documents: Monitoring center and audit logs
Open Integration
Capability: The product provides OpenAPI, API Key, and embedding capabilities for secondary integration.
Implementation: OpenAPI and API Key are provided for programmatic access. Embedding capabilities allow integrating data applications into existing portals; embedding does not expand the permission scope of the embedded object.
Benefits: Supports private deployment; ensuring data remains within the domain requires proper configuration of model, tool, and external service deployment and access. Organizations with existing portals can embed analytics capabilities into their systems.
Related documents: Open API, Embedding and Integration, Public Sharing
Interpretable Execution Process
Feature: AgentReplay provides a read-only interpretive view of retained sessions. The map displays the stable objects and clear relationships actually accessed in this request, while the trace shows the complete execution timeline. Abnormal calls are marked with steps that failed or showed "success but no available content".
Implementation: Playback only displays objects and relationships that can be determined from the actual input-output of real tools, without inferring from response text. Playback does not rerun or modify the original session.
Benefits: When issues arise, you can determine whether the cause lies within the data scope, tool invocation, execution steps, or result integration, avoiding repeated guessing based solely on the final answer.
Related documents: View and Verify Analysis Results