What this category's data looks like
This category's data mainly comes from CAD/CAM software, order BOM systems, sheet metal inventory ledgers, laser cutting and bending equipment, and MES systems. Daily documents include part unfolding drawings, CNC cutting programs, bending process cards, nesting diagrams, first-piece records, and process flow cards. Data updates are adjusted with each order, daily production, and each equipment model change. Most document fields include sheet metal specifications, bending coefficients, cutting dimensions, process flow nodes, and similar items. Units are mainly millimeters, square meters, kilograms, and man-hours.
What constraints these characteristics impose on the tool calling and plugins workflow
Data for this category is scattered across multiple systems and updated frequently. Tool calling plugins must support real-time synchronization of order BOM, equipment parameters, and inventory data. Daily documents include CAD unfolding drawings and process card text. Specialized plugins must be adapted to parse CAD formats and process text to extract key parameters such as bending coefficients and cutting dimensions, to avoid format parsing errors. The multi-variety, small-batch production mode requires tool calling to limit the recall scope to process documents associated with the current order, to prevent irrelevant data from interfering with Q&A accuracy. At the same time, process Q&A results directly affect production quality. Plugins must add a parameter verification link to avoid sheet metal waste or assembly deviations caused by incorrect parameters.
How to configure the setup
This page's configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Tool calling mode | toolChoice set to auto | Adapt to multi-scenario process Q&A, automatically select appropriate tools to perform parameter verification and data query |
| Number of recalled documents per single call | Top 3–5 | Avoid excessive recall causing context overload, align with the narrow-scope data requirements of multi-variety, small-batch production |
| Plugin timeout threshold | 600 seconds | Adapt to the time required for CAD file parsing and equipment data pulling, prevent process interruptions caused by slow operations |
| Document recall scope | Process documents associated with the current order | Limit data sources to avoid irrelevant nesting diagrams or historical process cards interfering with current Q&A |
| Function call format verification | Enable mandatory verification for functionCall | Ensure tool calling parameters comply with sheet metal process field specifications, such as the value requirements for bending coefficients |
Three common configuration mistakes
- Symptom: Slow operation logs are continuously reported during tool calling, and MongoDB response delay is too high. Cause: The document recall scope is not limited, so each call pulls all process documents, exceeding the system's processing capacity.
- Symptom: Q&A effects deviate significantly after tool calling is configured, and performance of models of the same level is worse than expected. Cause: No sheet metal process-specific verification is performed on tool calling parameters. For example, bending coefficient values are not combined with sheet metal thickness and material, leading to generated process parameters that do not meet actual needs.
- Symptom: Custom variable update plugins cannot implement counter auto-increment. Cause: The plugin is not configured to receive the current variable value as input, only supports fixed-value updates, and cannot read and accumulate existing values.
How to confirm the configuration is complete
- Initiate a process Q&A test, input "Current order's bending process card parameters", check whether tool calling only recalls documents associated with the current order.
- Trigger tool calling, view FastGPT system logs, confirm there are no abnormal delay prompts, and that records of
toolChoiceautomatically selecting the corresponding tool are normal. - Call the variable update plugin, manually pass test values, confirm the plugin can correctly receive and update variable values.
- Verify specialized tool calling, check that request parameters match the target service requirements, and confirm returned results conform to expected formats.
The values provided here are common starting points, and should be evaluated against the reader's own samples.