What the data for this category looks like
This category covers full-process quality-related records from work-in-progress to finished products, including control plans, patrol inspection records, SPC reports, finished product inspection reports, nonconforming product review forms, release forms, and monthly quality reports. Data primarily comes from MES systems, online inspection equipment, laboratory test results, process parameter libraries, customer quality standards, and historical defect knowledge bases. Data updates are triggered by production batches and team shifts, or collected continuously in real time by equipment. Single-batch data includes fields such as process number, inspection item, measured value, judgment result, defect type, and associated process parameters.
Constraints for Tool Calling and Plugins
Dispersed data sources and uneven update frequencies require tool calling to support parallel pulling of multi-source data, to avoid process slowdowns caused by delays from single data sources. Large differences in the structure of multiple document types require plugins to adapt to various formats including inspection records and SPC reports, and automatically extract standardized fields. The volume of historical defect library data accumulates over production cycles, so tool calling must support paged pulling to prevent overload from single requests. Data related to finished product release must be strictly associated with batch information. Plugins must automatically bind the batch ID as a query condition when called, to ensure the accuracy of data traceability.
Configuration Settings
This configuration is compatible with the plugin system of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Automatically match the plugins required for the current workflow, adapting to multi-source data pulling requirements |
| Workflow/API function-call configuration | strict: true | Force plugins to return results that conform to the preset format, avoiding missing fields or format errors |
| Plugin timeout duration | 600 seconds | Covers the reasonable processing time for multi-source data pulling and historical defect library queries |
| Batch association filter | Bind the current production batch ID | Ensure tool calling only pulls quality data for the corresponding batch, ensuring traceability accuracy |
| Parallel call limit | First 3 data sources | Control the volume of concurrent requests to avoid system overload |
| Single query return limit | First 1000 records | Limit the volume of data returned in a single request to prevent transmission overload |
Three Common Configuration Mistakes
- Phenomenon: A separately exported workflow plugin cannot be loaded in other workflows. Cause: The plugin configuration does not have global sharing permissions enabled, and is only bound to the context permissions of the original workflow.
- Phenomenon: The execution time of database connection plugins increases significantly, with notable differences between the new version and the old version. Cause: FastGPT v4.17.0 adds batch data association verification logic, and cross-data source consistency checks are enabled by default, leading to additional query overhead.
- Phenomenon: Intermediate steps of tool calling are directly displayed in the final output. Cause: The "Hide tool execution process" setting in the interface is not enabled, causing plugin call logs to be output synchronously.
How to Verify Correct Configuration
- Initiate a quality query for a simulated production batch. Verify that tool calling only pulls data for the corresponding batch, with no records from unrelated batches.
- View the plugin execution log. Confirm that the returned results conform to the format specified by
functionCall, with no missing fields or format errors. - Test parallel calling of multi-source data. Confirm that the system does not throw overload errors, and the execution duration meets the preset timeout threshold.
- Export the plugin from the current workflow. Attempt to import the plugin into other workflows, and confirm that it can be loaded and used normally.
The values provided here are common starting points, and should be evaluated against samples specific to the deployment.