What Data Is Included for This Category
Data for in-app natural language retrieval via function entries falls into two categories:
- Entry configuration data, sourced from the Function Entry configuration interface in the FastGPT admin backend. This includes fields such as
appId,apiSecret,bindDatasetIds, andsimilarityThreshold.bindDatasetIdsis an array type, andsimilarityThresholdis a unitless floating-point number. - User interaction data, sourced from user trigger logs after the app has been embedded. This includes fields such as
query(character unit),userId, andrequestTime(millisecond unit).
Configuration data takes effect in real time within 10 seconds after an administrator saves the settings. Interaction logs are synced to external systems in minute-based batches.
Constraints Imposed on HTTP Interfaces and External Systems
Function entry configuration data contains sensitive apiSecret, so external systems must use the HTTPS protocol for all interface requests to prevent key leaks. bindDatasetIds is an array-type field. When external systems configure entry-associated knowledge bases, parameters must be passed strictly in array format, otherwise retrieval scope will be abnormal. Interaction logs are synced in minute-based batches. External systems must support batch reception of logs via interfaces or configure a scheduled pull task every minute to ensure complete log data. similarityThreshold is a floating-point parameter. External systems must perform valid range checks on incoming threshold values to avoid retrieval results that do not meet expectations due to incorrect parameter formatting. Additionally, real-time requirements for in-app retrieval mean HTTP interface response delays must match user interaction wait expectations.
How to Configure Parameters
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
apiSecret | Generated separately for each terminal entry, stored in server environment variables | Prevents shared conversation records across users from public keys, complies with user data isolation requirements in financial scenarios |
maxRetrieveNum | Top 10 entries | Controls the number of returned results, avoids information overload in financial scenarios, and adapts to the simplicity requirements of in-app interactions |
similarityThreshold | 0.75–0.85 | Balances retrieval precision and recall, adapts to retrieval matching requirements for financial professional documents |
triggerMode | Triggered on input | Matches the real-time interaction habits of in-app natural language retrieval, improves user experience |
requestTimeout | 30 seconds | Adapts to the time consumption requirements of financial document parsing, avoids interrupting the retrieval process due to timeout |
batchSyncInterval | 60 seconds | Matches the batch reporting rhythm of interaction logs, ensures timeliness and completeness of log data synchronization |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require individual analysis, and it is recommended to test on in-house samples before finalizing settings.
Three Common Configuration Mistakes
- Failed to access associated docx documents when calling the interface. Symptoms include returning a
404 Not Founderror or empty response content. Causes include not uploading the document to the bound knowledge base and completing index synchronization, or incorrect document access path configuration. - Shared conversation records caused by using a public
apiSecret. Symptoms include historical conversations of different terminal users being visible to each other. Causes include not generating independent keys for each terminal entry, and not implementing user-level data isolation. - Invalid plugin API call input parameters. Symptoms include returning a
400 Bad Requesterror or custom input parameters not taking effect. Causes include not passing the array-typebindDatasetIdsparameter as required by documentation, or thequeryparameter format not meeting specifications.
How to Verify Successful Configuration
- Call the test interface with a simulated user query, check if the returned results include relevant content from the bound knowledge base.
- View the Function Entry configuration page in the FastGPT admin backend, confirm that parameters such as
apiSecretandbindDatasetIdshave been saved correctly and no abnormal prompts appear. - Check the external system logs, confirm that retrieval requests have been sent successfully and the returned status code is
200 OK. - Simulate queries from two different terminal users, confirm that their respective conversation records are not leaked or shared with each other.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.