What This Category's Data Entails
This category covers the full production process of bakeries, cookie manufacturers, instant noodle factories, and frozen pastry plants. Daily documents include formulas and feeding lists, production records, baking or steaming curves, metal detection records, cleaning and disinfection records, and factory inspection reports.
Data is sourced from raw material qualification certificates, ERP/MES systems, weighing equipment, ovens or frying lines, laboratory testing systems, and food traceability ledgers. Update frequency is per batch, per shift, or daily.
Data fields include material names, allergen labels, temperature curve parameters, test results, shelf life, and more. Units include degrees Celsius, kilograms, minutes, and other standard units.
Constraints on Tool Calling and Plugins
Frequent line changes and allergen control requirements for this category mean tool calling must quickly match the latest formulas and cleaning records. Multiple heterogeneous data sources require plugins to connect to different system interfaces and adapt to varying data formats.
Data updated per batch, shift, or daily means tool calling must prioritize pulling the most recent batch’s production and testing data, and avoid using expired information.
The ready-to-eat nature of these products places higher accuracy requirements on exception reports. Tool-generated reports must strictly link to the production and testing records of the corresponding batch, and must not mix data across batches.
Configuration Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Adapts to multi-type tool calling needs, covering scenarios such as formula query, exception report generation, and data pulling |
| Tool call timeout | 600 seconds | Adapts to the waiting time for pulling real-time device data like oven curves, and prevents call failures from slow device responses |
| Number of returns per single tool call | Top 3 entries | Matches the production data volume of a single batch, and avoids redundant information interfering with exception report generation logic |
| Maximum context length | 8000–12000 token | Adapts to the context needs of multiple production records and test reports, ensuring tool calls access sufficient reference data |
| Plugin data source filtering rule | Match by batch number | Aligns with batch-based data updates, ensuring pulled production and testing data belong to the currently processed production batch |
Three Common Misconfigurations
- When
toolChoiceis set tonone, the knowledge base selection entry is not available for tool calling. This configuration disables the knowledge base association logic for tool calling, and only allows direct calling of preset functions. - When calling the AntVchart MCP plugin, only code blocks are returned and no charts are rendered. This occurs because necessary chart rendering parameters are not configured, or the data source connected to the plugin does not return renderable structured data.
- Core fields of exception reports returned by tool calls are empty. This happens because the data source is not filtered by batch number, and expired production and testing data are called, preventing matching of the current exception’s batch information.
How to Verify Successful Configuration
- Send a simulated exception request, and confirm whether tool calling triggers preset formula queries or data pulling actions to verify the
toolChoiceconfiguration is active. - Review plugin call logs to confirm the returned data source is the current batch’s production and testing records, matching the batch filtering rule.
- Call the chart plugin to confirm renderable visualization results are generated, verifying the plugin’s data source and rendering parameter configurations are correct.
- Submit test data for multiple batches, and confirm tool calling only pulls documents for the corresponding batch to avoid mixing cross-batch data.
The values provided here are common starting points and should be measured against the reader's own samples.