What data for this category looks like
This category of data typically includes completion documents, configuration manuals, training records, and acceptance certificates. Data sources include project document libraries, configuration lists, meeting minutes, and test reports. It is organized by individual project, with batch updates delivered per phase or when the project concludes. Documents generally contain project identifiers, version numbers, archiving status, seal information, and similar details. Overall data accumulates gradually as the project progresses.
What constraints these characteristics impose on knowledge base retrieval and recall
This category of data is organized by individual project and updated in batches per phase or when the project concludes. Retrieval must strictly limit the search scope to avoid interference from cross-project data. Documents include version numbers and seal information, so recall results must match version consistency. A version field filter must be added during retrieval. Because data is updated in batches, the document set in the knowledge base must be synchronized regularly to ensure the retrieval scope aligns with the current project’s archived documents. Document integrity checks are time-consuming, so recall results must cover all necessary document types to avoid missing critical materials.
How to configure settings
This configuration applies to FastGPT v4.17.0. The following are recommended configuration items and their justifications:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches keyword and semantic relevance of project documents, avoids recalling invalid archived documents |
limit | Top 10 entries | Controls the number of documents recalled per request, adapts to the limited number of documents per single project |
searchMode | keyword + semantic | Combines keyword and semantic retrieval to cover professional terms and fixed fields present in project documents |
usingReRank | Enabled | Sorts recall results by relevance, elevates the priority of critical documents such as acceptance certificates |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands search terms to match professional terms present in project documents |
collectionIds | Document collection ID specified per project | Limits the retrieval scope to ensure only archived documents from the current project are searched |
Three common configuration errors
- Empty results or unexpected result counts returned by API calls | The
collectionIdsparameter is not configured correctly, and the document collection ID list for the target project is not specified, leading to retrieval scope that is too broad or too narrow. - Model connectivity verification failure displayed in the interface, with an error containing "request exception" | The App/Admin server environment variables
AIPROXY_API_ENDPOINT(root address of the AI Proxy service, for example http://aiproxy:3000, do not append /v1 or /chat/completions) andAIPROXY_API_TOKEN(AI Proxy administrator token, consistent with the AI Proxy server configuration and cannot be empty) are not configured correctly. This is a startup verification or service connection issue. Do not replace the service deployment configuration using workflow global variables, application API request headers, or the general model management page. - Garbled characters displayed in the knowledge base for uploaded CSV files | The file encoding format is not specified, leading to file parsing errors and garbled characters.
How to confirm configuration is correct
- Check the App/Admin server environment variables, confirm that
AIPROXY_API_ENDPOINT(root address of the AI Proxy service, for example http://aiproxy:3000, do not append /v1 or /chat/completions) andAIPROXY_API_TOKEN(AI Proxy administrator token, consistent with the AI Proxy server configuration and cannot be empty) are correctly configured. - Initiate a retrieval request, check if the returned results include the specified project’s document collection ID to confirm that the
collectionIdsparameter is effective. - Test the retrieval function, check if the relevance of the recall results aligns with the configuration of
similarityandlimit. - Review the retrieval results, confirm that critical documents such as acceptance certificates appear at the top after the re-ranking function is enabled.
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 2026-10-19.