Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Daily business documents for this category include departmental budget drafts, project application forms, performance target forms, personnel expense calculation forms, budget approvals, and similar items. Data sources include the budget management integrated system, departmental staffing system, prior-year final accounts, project library, and Ministry of Finance budget preparation notices. Data is updated on an annual basis, with full updates completed only during the new year’s budget preparation cycle. Document structures include structured tables and policy provisions. Fields include unit code, staffing number, project budget amount, performance target, and similar items. Units include yuan, ten thousand yuan, person, and similar items.
What constraints do these characteristics impose on the tool calling and plugins workflow
The data sources for this category involve multiple internal systems and official policy documents. Tool calling must adapt to multi-data-source connection requirements. The annual update frequency means the knowledge base does not require frequent synchronization. Full updates only need to be completed before the budget preparation cycle. Daily documents include structured table content. Tool calling must configure dedicated parsing plugins to extract accurate fields. The strong correlation between policy and budget data requires tool calling to cover both knowledge base retrieval and real-time data querying. It is also necessary to ensure the accuracy of recalled information, to avoid budget audit failures due to incomplete information.
How to set configurations
| Configuration Item | Recommended Value | Basis for This Setting |
|---|---|---|
toolChoice | auto | Supports automatic matching of tool calls and direct question answering, adapting to mixed needs for policy basis queries and budget data retrieval |
| Knowledge base recall limit | Top 8 entries | Departmental budget-related document scale is moderate; excessive entries will exceed context window capacity |
functionCall | Enabled | Requires calling dedicated tools to obtain real-time preparation data from the budget management integrated system |
| Structured document parsing switch | Enabled | Daily business documents include structured content such as personnel expense calculation forms and project application forms, requiring accurate field extraction |
| Context window length | 8000–12000 token | Must carry policy provisions, budget report data, and user questions simultaneously to ensure complete information |
Three common mistakes
- Issue: Insufficient number of returned results when calling the knowledge base, failing to cover all policy bases. Cause: The knowledge base recall limit is set too low, not matching the actual number of departmental budget documents.
- Issue: Uploaded PDF format budget documents cannot be correctly parsed into usable text. Cause: The PDF-to-structured text plugin is not configured, or the corresponding parsing rules are not enabled.
- Issue: Request timeouts or rejections occur during concurrent calls, or locally deployed large models cannot be called. Cause: Concurrency limit parameters are not adjusted, or network proxy settings for local large model calls are not configured.
How to confirm proper configuration
- Submit a test question that includes policy provision queries and budget data requests, check the tool calling log to confirm that
toolChoiceautomatically triggers the corresponding tool. - Upload a departmental budget draft document, verify that the parsing result extracts core fields such as staffing number and project budget amount.
- Submit batch test requests, confirm that there are no obvious timeouts or errors during concurrent calls.
- View the number of knowledge base recall results, confirm that they match the preset recall limit.
The values provided here are common starting points, and should be evaluated 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.