What the data for this category looks like
The core data for this category comes from cell grading and capacity testing equipment, manufacturing execution systems (MES), welding equipment, torque tools, BMS software, and EOL test benches. Daily documents include grouping rules, BOM lists, welding and torque records, BMS flashing records, insulation and voltage withstand reports, EOL test reports, and system configuration lists. Data is updated once per produced battery pack or energy storage cabinet, and synchronized whenever the BMS software version changes. Document fields include cell grading capacity and internal resistance, welding torque values (unit: N·m), insulation resistance values (unit: MΩ), BMS configuration parameters, and more. The size of a single document ranges from tens of KB to several MB.
What constraints these characteristics impose on the tool calling and plugins workflow
Data sources for this category are scattered across multiple independent devices and systems. Tool calling requires connecting to multiple types of data source plugins, and cross-system data pull logic must be configured. Data update frequency is high and tied to individual finished products, so tool calling must ensure real-time performance and avoid using expired cached data. Document structures are diverse and include professional parameters, so tools must support parsing structured BOMs, torque records, and unstructured test reports. Audit results are directly linked to product safety. Failed tool calls must trigger clear alerts and retry mechanisms to prevent incorrect data from entering subsequent stages.
How to set the configurations
The following configurations are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Adapts to multiple tasks including supplier audits and process quality inspections, automatically matching corresponding data source plugins |
| Workflow/API function-call configuration | enabled | Forces tool calling to avoid the AI generating unverified natural language responses directly |
| Plugin timeout period | 120–180 seconds | Covers cross-system data pull network delays and device response times, preventing call interruptions due to timeout |
| Number of retries on call failure | 2 times | Balances coverage of temporary network fluctuations and resource usage, avoiding excessive system load from repeated calls |
| Maximum parallel plugins | 3 | Matches the three core data source connection requirements for this category: MES, welding equipment, and BMS systems |
Three common mistakes
- Scenario: A database plugin call returns a connection failure, and the interface prompts "database connection timed out". Cause: The data source address and account permissions for the SQL Server plugin are not configured, and a timeout period suitable for the scenario is not set.
- Scenario: The AI does not trigger tool calls during the tool calling phase, and directly generates a natural language response. Cause: The
functionCallforced calling mode is not enabled, or the configured tool parameters do not match the current audit task. - Scenario: The document content associated with the audit result is empty, and the capacity grading data or test report for the corresponding cell cannot be matched. Cause: The association logic between plugins and the knowledge base is not bound, or the field mapping rules for document parsing are not correctly configured.
How to verify the configuration is complete
- Run a grouping rule query test for a single battery pack. After entering the corresponding instruction, check whether the plugin returns the cell grading data from the MES system.
- Trigger a simulated welding record audit test, and verify whether the tool calls the torque data collection plugin and returns torque values within the preset range.
- Simulate a BMS configuration error troubleshooting scenario, and confirm that the tool call returns the associated content of the corresponding flashing record and system configuration list.
- Trigger a simulated plugin call failure test, and check whether the preset failure alert and log recording logic is triggered.
The values provided are common starting points and should be measured against the reader's own samples.