This document is written for FastGPT v4.17.0.
What the data for this category looks like
Business data for this category primarily comes from formula systems, raw material batch ledgers, internal mixers, Mooney viscometers and rheometers, vulcanizers, and laboratory testing equipment. Daily documents include formula sheets, weighing records, mixing curves, rubber compound quick test reports, vulcanization process cards, mold usage records, and finished product performance reports. Corresponding records are generated for each batch of rubber material, each batch of raw materials, and each molding cycle. Individual records include fields such as rubber compound ratio (unit: phr), process parameters (temperature, time, pressure), equipment numbers, batch numbers, and test values.
What constraints these characteristics impose on tool calling and plugins
The high-frequency multi-source data updates of this category require tool calling to support millisecond-level response and high-frequency triggering, to avoid interfering with real-time production processes due to waiting delays. The multiple types of process and test documents require plugins to adapt to different data formats, such as mixing curve value extraction and structured parsing of process parameters. The full-link traceability requirement requires tool calling to associate multi-source data including raw material batches, equipment operation records, and finished product test results, to enable forward and backward traceability of defect issues. The irreversible nature of vulcanization processing requires tool calling to pull complete historical process data, and cannot rely on single real-time data to complete defect analysis.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Tool calling switch | Enabled | Enables tool calling capabilities to support production-related functions such as formula data query and process parameter verification |
| Single tool call timeout | 30–60 seconds | Matches the production interval per molding cycle, to avoid interrupting ongoing production processes due to timeout |
| Tool result concatenation method | Concatenate in call order | Ensures timeline consistency of traceability data, facilitating full-process restoration for defect analysis |
| Workflow/API tool selection | auto | Adapts to multi-source data query requirements, automatically selecting appropriate tools to complete data pulling and structured parsing |
| Plugin input parameter validation | Enable required field validation | Prevents tool calling failures caused by missing key parameters such as batch numbers and equipment numbers |
| Maximum tool call count | 3 | Limits the number of tool call rounds per conversation, preventing excessive occupation of production processes |
Three common configuration errors
- Issue: After tool calling completes, no results are displayed in the dialog box, and only partial results load when re-entering the session. Cause: Real-time push logic for tool results is not configured, causing results to only synchronize when the session is rebuilt.
- Issue: When using the code running plugin to write scripts, specified elements cannot be extracted from returned JSON data, and Python code running support is unavailable. Cause: The plugin's built-in code running environment only preinstalls JS dependencies, no Python runtime is configured, and quick configuration for JSON path parsing is not enabled.
- Issue: After inputting correct process data parameters when calling the basic chart plugin, the returned result is "none". Cause: Formatted numerical arrays are not passed as required, or required configuration items such as chart type and axis units are missing from the parameters.
How to verify correct configuration
- Initiate a test conversation that includes raw material batch query, and check whether the tool can pull weighing records and mixing data for the corresponding batch.
- After triggering tool calling, wait up to 30 seconds to check whether the dialog box displays tool return results in real time, without needing to re-enter the session.
- After setting
toolChoiceto auto, initiate a cross-data source query request, and check whether the tool automatically selects appropriate plugins to complete data parsing. - Input a tool calling request with required parameters, and check whether the plugin intercepts calling requests missing key information.
The values provided here are common starting points, and should be validated against the reader's own samples.