DataPilot is delivered in a private deployment (self-managed) mode by default. The platform is deployed in the customer's own environment, as indicated by the self_hosted parameter. The deployment environment only determines the default boundaries for the appearance and visibility of related controls; it does not change the permission model. Whether a user can access an object is still determined separately by account status, role, membership, and resource permissions. This topic focuses on deployment instructions, known specification limits, and capability boundaries of the current version, for reference during solution evaluation and troubleshooting, avoiding repeated searches across multiple chapters.
Notice
This topic lists only the limitations explicitly stated in the product documentation. The real-time upper limit prompts displayed on the page (for example, Current/Valid Upper Limit in the Agent data page title) take precedence over the default values in this topic, because the deployment administrator may have adjusted the configuration within the allowable capacity range.
Deployment and installation
For information about the supported installation methods and procedures, see Deployment using tools, High-availability deployment, and OAT deployment.
The Monitor in workspace settings observes only the current workspace.
Specifications and quotas
Upper limit for Agent data scope
Project |
Value |
Description |
|---|---|---|
| Number of data objects supported by a single agent (default) | 256 | The number of major compactions for tables, views, and external tables. |
| Maximum Number of Data Objects per Agent | 1000 | The deployment administrator can increase the default value based on the environment capacity, but the effective upper limit for any deployment will not exceed 1,000. |
The Data tab on the Agent configuration page displays Current/Valid Upper Limit in the selected list title, and the page prompts use the same valid value. When the upper limit is reached, the add button is disabled and an upper limit prompt is given.
The page limits are only for immediate feedback. The server is the final arbiter for creation, editing, workspace import, cloning, publishing, and automatic table addition for data applications. If a save, clone, publish, or automatic addition returns an out-of-limit error, the write is rejected and the original data scope remains unchanged. You should first bring the scope within the valid upper limit displayed on the page before retrying.
Notice
1000 is the hard cap for the number of table entries, which does not mean that table identifiers of any length can be saved unconditionally. Before increasing the deployment default value, administrators must also assess the actual storage volume.
For detailed operations, see Create and configure an Agent.
Agent runtime protection thresholds
Project |
Value range |
Default Value |
Description |
|---|---|---|---|
| Maximum Number of Inference Steps per Request | 1~1000 | 500 | When the upper limit is reached, the current round of dialogue will provide a prompt with the actual limit value, indicating that the request should be split or simplified. |
| Automatic Retry Attempts for Model Invocations | 0~5 | 2 | It only takes effect for transient errors such as timeouts, rate limits, and temporary service unavailability. |
These two items belong to the behavior settings of a single Agent and are retained with the Agent's cloning and published versions. The number of inference steps is a runtime protection threshold. Increasing it does not guarantee that complex tasks will definitely complete. See Behavior parameters and custom rules.
Files and uploads
Scenario |
Limitations |
|---|---|
| Chat Attachment | Files: up to 10 MB each. PDF, PPT, and PPTX files: up to 50 MB each. |
| Document Upload and OCR | The maximum size for a single PDF, PPT, or PPTX file is 50 MB. By default, OCR can process up to approximately 20 pages. |
| Feedback Attachment | Up to 5 files, with each file not exceeding 20 MB and the total size not exceeding 50 MB. |
| Voice input | A single recording can last up to 60 seconds, and the audio file size must not exceed 10 MB. |
| Batch Original File Upload | Exactly 100 MB is allowed. Exceeding this size will result in rejection before entering the long-duration processing. |
| Batch Inference of Classifiers | A maximum of 500 files at a time. PDF files must be single-page only. |
| Classifier Training Set | The maximum size of a ZIP file is 200 MB. After decompression, you can import at most 500 valid files. |
Other hard upper limits
Project |
Limitations |
|---|---|
| Running parameters of the notebook | Up to 32; text parameters up to 2,048 characters; up to 100 multiple selections |
| Workflow Scheduled Trigger | Each workflow can have at most one instance. The value range is as follows: minute interval: 1–59; hour interval: 1–24; day interval: 1–365. |
| Multi-dimensional Table Analysis Grouping Dimension | Up to 3 fields |
| Depth of Derived Tables | Multiple levels of derivation are supported, but circular dependencies are prohibited. |
| Subscription Application Batch Approval | Up to 100 records at a time |
| Agent Query Multidimensional Table Tool | A maximum of 200 rows are returned each time. |
| Report Review | At most two rounds of independent checks can be performed. |
| Feedback | Body content is required. Attachments are supported. |
License capacity
The license statement defines the scope of authorization, including the validity period, capacity, and purchased capabilities. When the capacity reaches its upper limit, you should first clean up or adjust the actual objects, and then update the license according to organizational procedures.
A valid license does not guarantee that the user has obtained all permissions: even if a feature is available, you must still check the user's permissions, service configuration, and operational health separately.
Concurrency and Queuing
The concurrency level for batch testing determines the maximum number of requests that can be processed simultaneously. Increasing concurrency may reduce waiting time, but it is still subject to the platform's overall capacity, model, and data source limitations. When a task is in the queue, no percentage progress is provided, and the task should not be resubmitted; instead, the system administrator should confirm the global capacity and worker health.
The RPM/TPM values on the model side serve as runtime admission protection, not proof of vendor package entitlement. The configured values must be consistent with the actual account capacity.
Dependencies and Compatibility
The platform's operation depends on the following external conditions. If any one of these conditions is not met, the related feature may remain unavailable even after a successful save:
Dependencies |
Purpose |
Owner |
|---|---|---|
| Data Source Connection | Metadata discovery, query execution, and persistent data writing | Space administrator configuration, network conditions provided by the system administrator |
| Model Service | Dialogue Reasoning, Semantic Processing, AI Columns, Report Generation, Judge Evaluation | system administrator |
| Sandbox and Runtime | Notebook execution, data application preview, and dashboard construction | System Administrator / O&M Engineer |
| Endpoints and External Services | Service addresses and authentication required for platform operation | system administrator |
| Observability Service | Monitor, Audit, System Logs, and Task Center | O&M |
A successfully saved form only means the configuration has been persisted, not that the entire workflow is established. All dependencies must be verified through corresponding business actions: data sources via test connections and data table preview, models via connectivity probing, and sandboxes and endpoints via a single real business execution.
Supported Data Source Types
The following types are explicitly listed in the product change log. The specific options depend on the currently deployed data source, as indicated by the drop-down list:
Type |
Hierarchical model |
Description |
|---|---|---|
| MySQL | Database; can configure an external catalog | After a catalog is configured, dialogues, semantic processing, workflows, and data applications all read metadata from this catalog to execute queries. |
| AnalyticDB for MySQL | Database | It uses the MySQL protocol on port 3306 by default. This type is licensed. |
| OceanBase Database in MySQL-compatible mode | The old version is Database. For versions that support external catalogs, both Catalog and Database layers are displayed. | External catalogs are available only when both the database version and user privileges meet the requirements. |
| OceanBase Oracle-compatible mode | Set the analysis scope to Owner. | Multi-dimensional tables support solidification in the same dialect; cross-dialect solidification additionally supports Oracle source to MySQL target, but not vice versa. |
| StarRocks | Catalog | Behavior is not affected by compatibility changes to the MySQL protocol. |
OceanBase materialized views are included in the table/view list and can be selected and queried directly.
Known Limitations
- Data Table Preview is Not a Full Table Export: Data table preview returns at most 10 rows; permission preview applies policies based on the selected preview identity, and external hosting mode follows the hosting configuration. The preview result is a bounded sample and cannot serve as proof of the complete data volume.
- Batch Actions in System Management Lack Independent Task Details: Batch actions such as importing users, generating or cleaning up activation codes, and importing licenses are request-based. Success is determined by the updated rows, quantities, or status. The task center does not provide detail pages for these requests.
- Task Center Cannot Modify Business Configurations: The task center only projects task facts from the source module and provides retry when allowed by the source task. Creation, editing, and data correction are still the responsibility of the respective business modules.
- Retries Do Not Overwrite Previous Attempts: A retry creates a new run attempt. The failure facts of previous attempts are retained for inspection and are not overwritten by the new progress.
- System Management Pages Do Not Provide SSO Pre-Binding Forms: External identity pre-binding must be completed by trusted external systems through the administrator integration interface.
- Aggregated Monitoring Metrics Are Not Real-Time Values: The health color, badge, and trend in Monitor are aggregated window signals and may have collection delays, which do not constitute a real-time service commitment.