What data for this category includes
Daily documents for sheet metal cutting and bending include unfolding drawings, CNC cutting programs, bending process cards, nesting drawings, first-piece records, and process flow cards. Data sources include CAD/CAM software, order BOMs, sheet metal inventory systems, laser cutters, bending machines, and MES systems. Update frequency follows per order, daily, or each time setup is changed. Core fields include sheet thickness (unit: mm), bending coefficient, cutting size (unit: mm), bending angle (unit: °), and others. Document structure and fields vary across different orders.
Constraints for tool calling and plugins
Due to the multi-variety, small-batch, drawing-driven nature of this category, the document set for each review changes with each order. Tool calling must support dynamically loading data sources for different orders. Diverse document types including CAD format files require plugins to adapt to parsing and calling different formats. Core review steps rely on external tools such as bending coefficient calculation, nesting verification, and bending sequence verification. As a result, tool calling must have flexible trigger and execution logic. Data comes from multiple systems, so plugins must complete cross-system data format conversion and synchronization to prevent review failures caused by format mismatches. Frequently updated data also requires tool calling to support near-real-time response capabilities.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Adapts to cross-review scenarios with multiple orders and multiple documents, automatically selects matching tools to complete reviews |
| Workflow/API function-call configuration | Enabled | Supports calling external plugins such as bending coefficient calculators and nesting verification tools to complete core review operations |
| Number of single tool calls | First 3 entries | Core review data for sheet metal cutting and bending is concentrated in the first few core documents; excessive entries will increase context load |
| Tool call timeout period | 60-120 seconds | Response times for connecting to CNC equipment and CAD tools usually fall within this range; a longer period will cause process lag |
| Context retention length | 800-1200 characters | Core review content for a single sheet metal process card falls within this range; excessive length will interfere with tool calling accuracy |
Three common mistakes
- Phenomenon: When attempting to establish cross-review links for documents without tools, the system prompts a configuration error. Cause: The
functionCallparameter is not enabled, or the call permission for the corresponding tool is not configured, making cross-document association without tools impossible. - Phenomenon: The output content includes the phrase "Reference Mark: [1]". Cause: The output switch for knowledge base reference marks is not turned off, or document reference identifiers are not filtered during tool calling.
- Phenomenon: Custom interface calls are not triggered after the user inputs a specified keyword. Cause: No keyword trigger rule is set in the plugin configuration, or the rule configuration is incorrect, causing the interception logic to fail to take effect.
How to confirm the configuration is complete
- Enter the plugin management page of FastGPT v4.17.0, confirm that the tools related to sheet metal cutting and bending have been enabled and configured with correct data source connection information.
- Input a test request containing core keywords such as sheet thickness and bending angle, observe whether the tool calling log has normal interface request and return records.
- Initiate a complete technical document cross-review test, check that the output content does not include unnecessary reference mark phrases.
- Input preset interception keywords, confirm that the system triggers custom interface calls and returns expected responses.
The values provided here are common starting points, and should be evaluated against the reader's own samples.