What data in this category looks like
Data for this category mainly comes from local public data open platforms, integrated big data platforms, departmental business databases, and electronic certificate repositories. Daily supporting documents include public data resource catalogs, data open lists, data item descriptions, desensitization rules, interface documents, and update frequency descriptions. Data updates occur monthly. Core fields include data item unique identifier, open usage scope, desensitization processing method, external interface address, and update cycle. Main formats are structured tables and interface metadata.
What constraints these characteristics impose on tool calling and plugins
The monthly data update requirement means tool calling must regularly sync the latest public data catalog and interface list. This prevents calling expired or newly added data items. Structured data from multiple sources means plugins must be compatible with different formats of interface metadata and desensitization rules. Plugins must also pre-verify the authorized usage boundaries of data. The variety of supporting documents means tool calling configurations must be linked to corresponding data item descriptions and interface documents. This ensures calling parameters match open usage conditions. Public data attributes require operation logs to be retained during tool calling. This supports tracing of data usage.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
toolChoice | Auto-select | Adapts to scenarios where the public data catalog includes multiple types of data interfaces, and automatically matches the calling method for corresponding data items |
functionCall | Enabled | Supports structured calling of public data interfaces, and matches parameter requirements in data item descriptions |
| Tool calling timeout | 300 seconds | Matches typical response delays of public data interfaces, and prevents call interruptions caused by large data volumes |
| Tool calling retry count | 2 times | Addresses temporary fluctuations in public data interfaces, and reduces invalid call attempts |
| Context window size | 8000 token | Supports loading multiple supporting documents including public data catalogs and interface documents, and ensures complete configuration information is available during calls |
Three common misconfigurations
- Symptom: Connection failure appears when calling tool services via SSE. Cause: No authorization credential configuration for public data interfaces is linked, and access parameters that meet open conditions are not included.
- Symptom: Links cannot automatically recover after tool calling stops. Cause: No retry mechanism for tool calling is configured. Only single-call logic is used, and temporary fluctuations in public data interfaces are not accounted for.
- Symptom: Large language model output includes the phrase "citation mark: [1]". Cause: Structured citation mark output is not disabled in tool calling configurations. Reference information from the public data catalog is directly spliced into the output.
How to confirm configurations are complete
- Run a tool calling test, and verify that returned data items match configurations in the public data catalog.
- View tool calling logs, and confirm that corresponding data item descriptions and desensitization rule configurations are linked.
- Trigger multiple calls, and verify that the retry mechanism functions normally during interface fluctuations.
- Check large language model output, and confirm that unexpected structured citation marks do not appear.
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.