What this category of data looks like
Data sources for this category include design BOM, silicon steel sheet and insulating material data, vacuum and drying equipment, oil testing, and test stations. Data is updated per product unit and per batch of materials. Daily documents include winding and core assembly records, vacuum drying records, insulating oil reports, partial discharge and withstand voltage test reports, and nameplate parameter tables. Document fields include insulation distance, moisture content, partial discharge value, rated capacity, voltage level, and other relevant metrics, with units matching industry standard units such as millimeters, percentage, picocoulomb, kilovolt, and others.
What constraints these characteristics impose on the tool calling and plugins workflow
The exclusive data volume for a single product in this category is large, and it must be strictly bound to single-unit traceability information. Tool calling must avoid cross-product data confusion. Data is updated in real time per product unit and per batch, so plugins must support on-demand pulling of the latest document content such as insulating oil reports and partial discharge test reports. Most documents are structured test records, so tool calling must adapt to field extraction and verification logic to ensure that parameters in tender documents match actual test data. Since errors may lead to serious issues such as insulation breakdown and power grid faults, a secondary verification link must be added to tool calling results to check whether parameters meet the compliance scope required for tendering. At the same time, the loading volume of single-unit data must be reasonably limited to avoid call timeouts.
How to configure the settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | ["function"] | Force tool calling to avoid the model defaulting to outputting knowledge base content |
| Workflow/API function-call configuration | Trigger automatically by product ID | Adapt to on-demand calling of single-product data |
| Tool call timeout | 300–600 seconds | Adapt to the parsing and pulling duration of single-unit test reports |
| Plugin data filtering rules | Only load insulating oil reports and test data for the current tender product | Avoid cross-product data confusion |
| Number of returns per tool call | First 5 entries | Control the volume of returned data to avoid exceeding context limits |
| Plugin context limit | First 8000–10000 characters | Limit the loading volume of single-product data to avoid call timeouts |
Three common misconfigurations
- Symptom: The model defaults to outputting knowledge base content after a call, and tool calling is not triggered. Cause:
toolChoiceis not configured to forced trigger mode, causing the model to prioritize using existing knowledge base content. - Symptom: Fields returned when pulling insulating oil reports via tool calling are empty. Cause: No plugin data filtering rules are configured, causing irrelevant batches of material data to be loaded.
- Symptom: When calling the application via API, single-product parameters cannot be assigned. Cause: No product ID is bound as a global variable input parameter in the plugin configuration, causing parameters to fail to be passed correctly.
How to confirm configurations are correct
- Trigger a tender document generation request, check the tool calling logs, and confirm that
functionCallis triggered automatically. - Check the parameters returned by the tool call, confirm that only insulating oil reports and test data for the current tender product are included.
- Simulate an API call, pass a test product ID, and confirm that global parameters can be assigned correctly and return corresponding data.
- Check the tool call timeout logs, confirm that no timeouts exceeding 300–600 seconds occur.
The values provided here are common starting points and should be evaluated against the reader's own samples.