What data for this category looks like
The data for this category comes from engineering orders, sand, gravel, cement and admixture inspections, moisture content tests, mixing plant control systems, and project laboratories. Daily documents include mix proportion notices, opening identifications, production and transport slips, raw material inspection reports, test block records, and strength reports. Data updates are synchronized with each vehicle’s production, each batch of raw material deliveries, and daily laboratory tests. Core fields include aggregate moisture content, mix proportion parameters, slump requirements, and test block strength values, which are circulated in real time throughout production and quality processes.
What constraints do these characteristics impose on the tool calling and plugins workflow
Data updates for this category are synchronized with each vehicle’s production, each batch of raw material deliveries, and daily tests. The high-frequency, batched update requirement means tool calling must support real-time pulling of production and quality data for the corresponding batch. Multiple types of business documents have differentiated fields, so plugins must adapt parsing logic for different fields such as mix proportion parameters, aggregate moisture content, and slump requirements. Single-batch production data is independently associated with engineering orders, so tool calling must bind the corresponding business identifier to avoid data confusion. Test block strength reports have latency, so plugins must support asynchronous pulling of verification results and association with the corresponding pouring batch.
How to configure settings
The following configurations are based on FastGPT v4.17.0
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Adapts to the automatic tool calling logic for process specification and technical standard Q&A, triggering corresponding business data pulling as needed |
| Workflow/API function-call configuration | {"name": "pull_concrete_data", "parameters": {"batch_id": ""}} | Q&A for this category requires binding to a specific production batch, dynamically passing the batch number to pull corresponding mix proportion and test data |
| Session context retention duration | 3600 seconds | The process from batching to pouring for a single batch of concrete usually takes within 1 hour. Retaining context for this duration allows association of historical data for the complete production cycle |
| Tool call result cache duration | 60 seconds | Aggregate moisture content is updated in real time with each vehicle’s test. The cache duration matches the test frequency to avoid using outdated data |
| HTTP request retry count | 2 times | Mixing plant control systems occasionally experience network fluctuations. Limited retries can ensure the success rate of data pulling |
| Tool parameter validation rules | Validate the matching relationship between aggregate moisture content, water consumption, and slump thresholds | The core risk of errors for this category comes from parameter mismatches. Pre-validation can reduce the probability of material returns and shutdowns |
Three common mistakes
- Symptom: Numeric parameters output by plugins cannot be passed in HTTP request components, and logs show that parameter fields are empty. Cause: The parameter type and mapping rules are not declared in the tool calling configuration, causing plugin output to not be correctly bound to request parameters.
- Symptom: When input questions contain the word "share", they are replaced with "hand over" when passed to tool calls. Cause: The word segmentation rules for input preprocessing do not adapt to the boundaries between professional terms and common words, leading to incorrect character splitting and replacement.
- Symptom: The output of the tool calling module is forcibly returned to the conversation interface and cannot be hidden. Cause:
toolChoiceis not configured asnone, or the switch for automatically returning tool results is not turned off, causing call results to be displayed directly.
How to verify correct configuration
- Submit a test question containing the keywords mix proportion and slump, check whether tool calling is triggered and whether production data for the corresponding batch is pulled.
- Simulate a single-batch production process, check whether the tool calling context retains the associated information of the corresponding engineering order and vehicle batch.
- Trigger a tool call failure scenario, check whether the specified number of retries is executed according to the configuration and corresponding error information is returned.
- Verify that the output of tool calling can be configured to be hidden or displayed as needed, matching business scenario requirements.
The values provided are common starting points and should be measured against the reader's own samples.