What data for this category looks like
The data for this category primarily comes from the order and inventory modules of the enterprise ERP system, BOM lists, process route files, equipment capacity ledgers, staff scheduling sheets, work-in-progress inventory records, and supplier arrival notices. Daily processed documents include sales orders, master production schedules, material requirements plans, weekly and daily production scheduling sheets, production work orders, material shortage lists, and delivery change notices. Data update frequencies include daily, weekly, and real-time updates when orders are released or exceptions occur. Data fields include work order numbers, expected process durations, equipment IDs, material codes, work-in-progress quantities, etc., with corresponding units such as hours, pieces, tons.
What constraints these characteristics impose on the tool calling and plugins workflow
Multi-source heterogeneous data sources require tool calling to support integration with ERP, equipment ledgers and other data sources in different formats, to avoid data parsing deviations. The high-frequency and irregular update rhythm requires tool calling configurations to support on-demand triggering, to ensure the timeliness of ledger data. The diverse document types and field differences require plugins to preset standardized field mapping rules, to adapt to the parsing logic of different business documents. The real-time update requirement in exception scenarios also requires the tool calling chain to have rapid response capabilities, to reduce delays during rescheduling or summary.
How to set up the configurations
The following configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API function-call configuration | required | Production ledger summary requires mandatory tool calls to pull real-time ERP and equipment data, to avoid the large language model generating fabricated content |
| Tool call timeout threshold | 600 seconds | Pulling production data involves integration with multiple systems, so sufficient response time must be reserved to cover typical cross-system interaction delays |
| Streaming return parsing rule | Split and merge by the detail field | In this scenario, the detail field returned by the tool contains core ledger data such as processes and materials, which must be fully extracted and summarized to generate reports |
| Knowledge base recall count | Top 8 entries | An appropriate number of process and scheduling knowledge related to production planning should be recalled, to avoid context overload affecting tool call judgment |
| Workflow/API tool selection | auto | In conventional scenarios, the large language model independently determines whether to call tools. For emergency order insertion scenarios, it can be manually adjusted to mandatory call |
| Online tool enablement status | Enable as needed | Some delivery change and supplier arrival data requires online supplementation, which is not required for all scenarios, to reduce unnecessary call costs |
Three common misconfigurations
- Symptom: The
detailfield returned by tool calls is empty, and the generated ledger data lacks process and material information. Cause: The streaming return parsing rule is not configured, and the logic to extract thedetailfield is not specified. - Symptom: Frequent timeout errors occur during tool calls, with a 504 status code. Cause: The tool call timeout threshold is set to a value lower than the actual system response duration, such as
300 seconds, causing cross-system data pulling to be terminated before completion. - Symptom: Some tools cannot be triggered in Docker-deployed applications, and the tool call judgment logic does not work as expected. Cause: Network ports required for tool calls are not opened in the container configuration, preventing the application from connecting to external ERP or equipment systems.
How to verify the configuration is correct
- Submit a query that includes a specific production work order number, and check whether the tool call successfully pulls the corresponding work order data from the ERP system.
- Enable streaming return testing, and check whether the returned result fully includes core information such as processes and materials in the
detailfield. - Submit a query simulating an emergency order insertion scenario, and check whether the tool triggers real-time data pulling and updates the content of the scheduling report.
- View the tool call logs, confirm that the timeout threshold setting matches the actual system response duration, and there are no frequent 504 timeout errors.
The values provided here are common starting points, and should be evaluated against the reader's own samples.