What the data for this category looks like
Reference only; verify against the target FastGPT release and deployment documentation.
What constraints these characteristics impose on the tool calling and plugins workflow
Since all data fields are plain text with no nested structure, no additional parsing of complex nested structures is required when calling tools, and specified fields can be directly extracted for recall matching. Due to the high update frequency and support for incremental synchronization, the cache validity period for tool calls must be set to no more than 1 hour to avoid returning outdated reply templates. The clear length range of a single piece of data requires limiting the total number of characters recalled in a single call when using tool calls, to prevent exceeding the model's context window limit. The existence of request classification tags requires that tool calls can filter recall results by classification tags to improve matching accuracy.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
toolChoice | auto | Adapts to the diverse classification of hotline requests, automatically matching the corresponding reply template recall tool |
functionCall | [{"name":"recall_reply","parameters":{"query":"example value","category":"example value"}}] | Accurately recall reply templates in the corresponding field based on the request classification tag submitted by the user |
| Cache validity period | 30 minutes | Matches the data incremental update frequency, balancing real-time performance and interface call efficiency |
| Single recall limit | Top 3 entries | The length of a single template is moderate, and the total length of 3 entries meets the model's context window requirements |
| Interface timeout period | 60 seconds | Adapts to the fast response requirements of hotline scenarios, avoiding excessive user waiting time |
| Cross-origin allow configuration | Add the hotline platform domain name to the allow list | Allow third-party systems to call FastGPT open interfaces via browsers |
Three common mistakes
- An error indicating the model does not support tool calling is returned during a call. The cause is that the model type supporting
functionCallortoolChoicewas not selected. - A 403 permission error is returned when calling the interface via a third-party browser. The cause is that the domain name of the hotline platform was not added to FastGPT's cross-origin allow configuration list.
- The recalled reply template does not match the user's request. The cause is that the classification tag of the user's request was not passed in the tool call parameters, resulting in an overly large recall range.
How to verify successful configuration
- When the tool interface is invoked, check whether the returned result contains the recall parameters corresponding to
functionCallortoolChoice, and confirm that the parameter format is correct. - When user requests of different classifications are simulated, invoke the tool interface, and check whether the recalled reply templates match the corresponding classifications, confirming that the filtering logic takes effect.
- Review the interface call logs, confirm that configuration items such as timeout period and cache validity period match the preset values, and no abnormal errors are present.
- When a browser call is initiated from a third-party system, confirm that the interface can return results normally, and no cross-domain related errors are present.
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.