What data for this category looks like
Data sources for this category include EAM/CMMS systems, equipment sensor data, operator inspection records, maintenance work orders, spare parts inventory lists, and statutory inspection records for special equipment. Update cycles cover shift-based, daily, weekly, monthly, and immediate updates during sudden equipment failures. Daily documents include equipment ledgers, inspection forms, graded maintenance plans, maintenance work orders, failure analysis reports, spare parts lists, and special equipment inspection records. Document fields include equipment number, model, installation date, inspection points, status markers, spare part codes, inventory quantity, inspection validity period, and more. Units include units, pieces, hours, days, and more.
What constraints do these characteristics impose on tool calling and plugins?
Multi-source heterogeneous data sources and varied update cycles require tool calling to support batch pulling of data sources across different cycles, and handle synchronization between real-time sensor data and offline ledgers. Rich document fields and associated business logic require plugins to accurately match fields related to equipment, spare parts, and maintenance, to avoid missing critical parameters. Errors in maintenance work directly impact production and compliance, so tool calling must include built-in data validation logic to prevent incorrect data from being included in bid quotations. Additionally, maintenance windows must align with production scheduling, so tool calling must link to production scheduling data to adjust quotation timelines, ensuring quotations match actual project execution conditions.
How to Set Up Configurations
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Allows the model to independently select tools matching bid requirements |
| Workflow/API function-call configuration | ["get_device_info", "get_spare_part_cost", "calculate_maintenance_quote"] | Binds three types of tools: equipment ledger query, spare part cost retrieval, and maintenance quotation calculation |
| Data source filtering rules | Filter by equipment workshop and inspection cycle | Matches production areas and maintenance requirements corresponding to the bid project |
| Field validation rules | Validate required fields: equipment number, spare part inventory quantity, inspection validity period | Prevents quotation deviations caused by missing critical data |
| Timeout period | 300 seconds | Adapts to reasonable response durations for multi-source data pulling and cost calculation |
| Global variable binding | Bind bid project number and project timeline | Uniformly passes context parameters to all tool calls |
Three Common Mistakes
- Phenomenon: Extra 0 values appear in quotation results returned by tool calls. Cause: No output filtering rule configured for
functionCall, and the model accidentally mixes irrelevant intermediate calculation values into the final text. - Phenomenon: Non-predefined reply content is automatically generated after tool call execution. Cause:
toolChoiceis not locked to the specified tool set, and the model independently triggers general question-and-answer logic. - Phenomenon: Spare part inventory data returned by tool calls does not match actual values. Cause: Data source synchronization frequency is not configured according to inspection update cycles, and expired inventory data is retrieved.
How to Confirm Proper Configuration
- Navigate to the FastGPT tool configuration page, confirm that
toolChoiceis set toautoor the specified tool set, and that parameter configurations match preset rules. - Initiate a simulated tool call, enter the equipment number corresponding to the bid project, and check if the returned equipment ledger data includes required fields.
- Trigger a tool call, then check if the output result contains extra irrelevant numerical values or non-predefined reply content.
- View the data source synchronization log, confirm that data updates at the preset frequency, and there are no records of expired data calls.
The values provided here are common starting points, and should be measured against the reader's own samples.