What the data for this category looks like
Core data for this category comes from formula systems, raw material batch management, internal mixers, Mooney and rheometer equipment, vulcanizing machines, and laboratory testing equipment. Daily documents include formula sheets, weighing records, mixing curves, rubber material rapid test reports, vulcanization process cards, mold usage records, and finished product performance reports. Data is generated once per batch of rubber, per raw material batch, and per mold run of products. Fields include business parameters with clear units such as feeding weight (kg), mixing temperature (℃), vulcanization time (min), torque (N·m). Some documents are multimodal formats with embedded curves.
What constraints these characteristics impose on the tool calling and plugins workflow
High-frequency updated data requires tool calling to support scheduled pulling at minute-to-hour intervals, to avoid data lag that impacts real-time ledger summary. Multiple types and multimodal documents require plugins to adapt to API formats of different data sources, including structured tables and unstructured curve images. Business fields with fixed units require automatic parameter format verification during tool calling, to prevent invalid data from entering the final report. Irreversible vulcanization processes mean ledger data cannot be modified. Tool calling must be configured with strict permission and verification logic to avoid data tampering.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Automatic tool selection adapted to multiple types of ledger data, covering formula, mixing, vulcanization and other links |
| Multimodal vector model configuration | Tongyi multimodal vector adaptation model | Supports vectorization processing of mixing curves and process card images |
| Tool call timeout period | 300–600 seconds | Adapts to delay fluctuations in data pulling from internal mixer and vulcanizing machine equipment |
| Workflow/API function-call configuration | Enable mandatory verification | Ensure that ledger fields match business formats and avoid invalid data summary |
| Cross-domain access configuration | Allow specified device API domain names | Support third-party browsers to directly call production equipment data sources |
| Failure retry strategy | Retry 2 times after failure, interval of 10 seconds | Respond to temporary network fluctuations in high-frequency data pulling |
Note: This configuration is based on FastGPT v4.17.0
Three Common Misconfigurations
- Phenomenon: An error indicating unsupported format is returned when calling the Tongyi multimodal vector model. Cause: The multimodal vector adaptation model is not specified in the tool configuration, and the default call only supports text-only vector models.
- Phenomenon: Tool calling does not select the specified tool for execution as expected, returning irrelevant results. Cause: The
toolChoiceparameter is not set correctly, or the trigger rule of the corresponding tool is not bound to the ledger type association. - Phenomenon: Cross-domain interception error is returned when a third-party browser calls the device API. Cause: The allowed third-party domain name whitelist is not configured on the FastGPT platform, and cross-domain access permissions are not enabled.
How to Verify the Configuration Is Complete
- Enter the tool management page, check whether the bound data sources match the category data sources (internal mixers, vulcanizing machines, laboratory equipment, etc.).
- Trigger a single ledger data pull, and check whether the returned fields include category-specific business parameters such as mixing temperature and feeding weight.
- Simulate the trigger scenario corresponding to the ledger type, and check whether the preset tool can be called correctly and generate report fragments that meet the format requirements.
- View the platform access logs, confirm that the cross-domain configuration has taken effect, and there are no interception records for third-party API access.