What the data for this category looks like
This category's data comes from raw material and auxiliary material qualification certificates, ERP/MES systems, weighing equipment, ovens/frying lines, laboratories, and food traceability ledgers from bread and pastry, biscuit, instant noodle, and frozen pastry factories. Daily documents include formulas and feeding lists, production records, baking/steaming curves, metal detection records, cleaning and disinfection records, and factory inspection reports. Data update frequency is per batch, per shift, or daily. Document fields include material codes, temperature parameters, test results, shelf life, and more, with units including degrees Celsius, kilograms, batch numbers, etc.
What constraints these characteristics impose on the tool calling and plugins workflow
The multi-formula and frequent line change characteristics of this category require tool calling to support quickly switching to exclusive knowledge base documents associated with different equipment. The frequent need for allergen and cleaning switching requires tool calling to accurately match exclusive fields of cleaning and disinfection records and allergen materials. The rhythm of real-time data updates by batch and shift requires tool calling configuration to support real-time pulling of the latest production and test data. The ready-to-eat product attribute requires tool calling results to be associated with metal detection and microbial detection fields to avoid safety risks. The differences in parameter formats across different production equipment require tool calling to adapt to the field structures of multiple data sources to avoid format mismatches.
How to Set the Configuration
This configuration guide is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale for This Setting |
|---|---|---|
| Workflow/API tool selection | Auto-adapt | Adapts to the tool calling scenarios of multiple equipment and multiple documents for this category, automatically matching tools in the corresponding knowledge base |
| Workflow/API function-call configuration | Enabled | Supports accurate calling of equipment inspection and maintenance related documents and production data interfaces to meet scenario requirements |
| Number of tool calls returned per single time | Top 3 | Matches the characteristics of single-batch production data volume for this category, avoiding redundant information interfering with core decision-making |
| Context window length | 800–1200 characters | Adapts to the document length of formulas and production records, ensuring complete key parameters and operation steps |
| Tool call timeout | 600 seconds | Adapts to the time consumption requirements of pulling data from multiple sources (ERP, ovens, laboratories), avoiding call interruptions |
| Tool call retry count | 2 times | Adapts to failure scenarios of pulling data from multiple sources, avoiding direct process termination from a single failed call |
Three Common Mistakes
- Application configuration tests complete normally, but API calls return errors. The cause is failure to synchronize tool calling configuration to the API key permission scope of the deployment environment.
- After upgrading the version, equipment-related knowledge base models cannot be called correctly. The cause is failure to update the adaptation configuration of
functionCall, leading to invalid field matching. - Tool calling parses production records too slowly. The cause is failure to limit the number of documents returned per single call, with excessive batches of production data loaded causing delays.
How to Confirm the Configuration Is Correct
- Simulated API calls can be initiated, and returned results are checked for the baking curve and allergen marking fields for the corresponding batch, confirming that the tool calling retrieval scope is correct.
- Tool calling logs are reviewed, and the number of returned documents is confirmed to meet the configured single return limit, with no redundant data loaded.
- Knowledge base documents for different production equipment are switched, and tool calling is tested for automatic adaptation to different field formats, confirming that the
toolChoiceconfiguration takes effect. - Timeout configuration is verified, and calls that pull data from multiple sources are initiated, confirming that no timeout interruptions occur and that the time consumption meets business scenario requirements.
The values provided are common starting points and should be measured against the reader's own samples.