Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
This category of data comes from the budget management integrated system, centralized treasury payment system, accounting system, and asset management system. The update cycle is annual. Daily produced documents include department final account statements, balance sheets, income and expenditure final account statements, final account explanations, and government financial reports. The document structure covers core fields such as revenue and expenditure details, asset and liability details, and budget execution difference explanations. Units mostly use yuan or ten thousand yuan, and must strictly follow unified fiscal caliber specifications.
What constraints do these characteristics impose on the "tool calling and plugins" link?
The annual update cycle requires that tool calling be limited to the dataset for the corresponding year, to avoid cross-cycle data confusion. Mixed input of multiple types of financial documents requires plugins to support both structured report field extraction and semantic parsing of unstructured explanatory text. The multi-system data source feature requires tool calling to connect to interface permissions of specified systems to ensure data source compliance. The high error risk in the final account link requires that tool call results include data traceability information to facilitate subsequent verification and audit.
How to configure the settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
toolChoice | auto | Adapts to the model's need to independently judge tool calling timing, aligns with the business scenario of on-demand data interface calls in department final account Q&A |
functionCall | ["query_financial_data", "verify_policy_basis"] | Explicitly specifies core tool types, covering the two key links of data query and policy basis verification |
| Single tool call timeout | 120 seconds | Financial system interfaces typically require a certain response time; 120 seconds covers most data retrieval scenarios |
| Number of recalled results | Top 8 | There are many policies and data entries related to department final accounts; 8 entries cover the core reference scope |
| Workflow tool trigger rule | Trigger based on policy matching degree | Matches the business logic of verifying policy basis first before calling data in department final account Q&A |
Three common mistakes
- Phenomenon: Tool calling cannot parse Word-format final account explanation documents. Cause: No Word-format document parsing plugin is configured, resulting in failure to correctly extract unstructured text.
- Phenomenon: Tool calling logic in the workflow does not meet expectations; the model does not independently select tools. Cause:
toolChoiceis not configured toautomode, resulting in a mismatch between tool calling rules and business processes. - Phenomenon: No notification response occurs after configuring the DingTalk webhook plugin. Cause: Correct interface signature and permission verification parameters are not configured, resulting in the system being unable to connect to the DingTalk notification channel.
How to confirm the configuration is complete
- Initiate a test query containing "202X annual department final account policy basis", and check whether the model triggers the preset tool calling action.
- View the tool calling logs in the FastGPT backend, and confirm that the returned financial data comes from the interface of the corresponding year's budget management integrated system.
- Simulate a data missing scenario, and check whether the tool calling triggers reasonable error prompts and retry logic.
- Send a test request to the configured DingTalk webhook address, and confirm that the notification channel can normally receive messages.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.