What Data for This Category Looks Like
Data for this category originates from design documents, standard instrument and environmental monitoring data, calibration software, laboratory information systems, and customer range requirements. Daily generated content includes assembly and commissioning sheets, calibration procedures, original calibration records, error curves, metrology certificates, and software version records. Data is updated per device and per calibration flow, and includes fields such as tested piece parameters, standard instrument readings, ambient temperature and humidity, error calculation values, and traceability numbers. Units include metrology-specific values such as pascals, degrees Celsius, and milliliters per minute. Calibration records and metrology certificates are core deliverables.
Constraints Imposed by This Category’s Data Characteristics on Tool Calling and Plugin Workflows
This category’s data characteristics impose clear constraints on tool calling and plugin workflows. Per single calibration update requirements, tool calling must bind the single-device single-calibration context to avoid cross-device data confusion. Core deliverables are metrology certificates with traceability numbers, so tool calling must strictly validate field completeness to ensure no missing key items including traceability information, environmental parameters, and error calculation values. Metrology-specific fields and units require tool calling to support unit parsing and format validation, preventing unit conversion errors. Requirements for acquiring data from multiple sources necessitate configuring stable call timeouts and retry mechanisms to address delays from multi-system interactions.
Configuration Settings
The following configurations are written for the tool calling module in FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Adapt to multi-tool calling scenarios, automatically match tool types required for process document drafting and exception report generation |
| Workflow/API function-call configuration | enabled | Enable tool calling functionality, support interfacing with multiple data source APIs such as calibration systems and standard instrument data |
| Single tool call timeout | 300 seconds | Address delays from multi-system acquisition of calibration data, avoid process interruption due to timeout |
| Tool call retry count | 2 times | Balance call success rate and process duration, adapt to occasional exceptions in multi-data source interactions |
| Field validation threshold | First 12 core fields | Ensure no missing key information such as traceability numbers, environmental parameters, and error values for metrology certificates and calibration records |
| Context window limit | Single device single calibration data | Prevent cross-device data from being included, ensure drafted content matches the current calibration task |
Three Common Configuration Mistakes
- A getaddrinfo ENOTFOUND error occurs when calling calibration system APIs. The cause is incorrect configuration of the target domain name resolution for tool calling, or an incorrectly entered target API address, resulting in failed connection establishment.
- A 400 status code is returned by the large language model when calling MCP tools, but normal operation occurs when MCP tools are not called. The cause is that parameters for MCP tool calling do not match the input format of the large language model, preventing the model from correctly parsing the tool calling request.
- An aiPointsNotEnough error is returned by the API call. The cause is that calibration data fields required for tool calling were not fully acquired, leading to insufficient necessary information for generating process documents or exception reports.
How to Confirm Configuration Is Correct
- Use the tool calling test function in FastGPT v4.17.0, initiate a tool calling test for single device single calibration data, and verify that the response format returned by the API matches expectations.
- Trigger the field validation logic, manually remove one core field, and confirm that the tool calling triggers the corresponding validation prompt.
- Adjust the tool call timeout time, simulate a delay scenario, and check whether the retry mechanism triggers as configured.
- Switch between different large language models, and confirm that the
toolChoiceandfunctionCallconfigurations can properly adapt to the model's tool calling rules.
The values provided are common starting points and should be measured against the reader's own samples.