What data for this category looks like
This category of data comes from cloud audit logs, IAM permission systems, CMDB resource inventories, and security baseline scan results. Daily produced documents include compliance checklists, permission audit reports, resource inventory lists, and rectification tracking forms. The data update cycle is monthly or quarterly. Document fields include resource ID, permission configuration items, compliance status, rectification responsible person, scan time, and other relevant fields. There are no unified units across fields. Some fields must be generated by combining data from multiple systems. The overall format is a mix of structured and semi-structured data, with a core focus on attributes related to compliance verification.
Constraints on knowledge base retrieval and recall
Data for this category comes from multiple dispersed systems. Retrieval tasks require associating cross-source fields, which raises higher requirements for matching accuracy of recalled fields. The monthly or quarterly update frequency means the knowledge base does not need high-frequency incremental synchronization. However, single imports of full resource inventory and other document entries are large in volume. This requires adapting to segmentation rules for long documents. Compliance documents have strong attribute correlation. Retrieval must accurately match core fields such as permission configurations and compliance status to avoid interference from irrelevant results. Rectification tracking forms must be linked to the latest scan data. The time range of recalled data must be limited to ensure results align with current compliance status.
How to configure settings
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets the precise matching needs of compliance documents, filters out low-correlation permission or configuration item results |
limit | Top 8–12 entries | Resource inventory documents have a large number of entries. Controls the number of recalled entries to avoid context overflow, while covering enough compliance verification dimensions |
searchMode | vector + keyword | Combines vector semantic matching and keyword matching, balancing precise recall of structured fields such as permission configuration items and semantic matching of compliance descriptions |
usingReRank | Enabled | Reranks recalled results to prioritize rectification tracking forms or permission audit reports with higher relevance to compliance issues |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands user compliance-related queries to cover more similar compliance documents |
collectionIds | Specified by compliance document collections | Only retrieves knowledge base collections related to cloud resource compliance audit, avoiding interference from documents of other categories |
Three common misconfigurations
- Calling the knowledge base retrieval API returns an unauthorized error, even after configuring the API key. The cause is that the dedicated knowledge base collection for cloud resource compliance audit is not bound to the corresponding API key, or the
collectionIdsconfiguration is incorrect, resulting in an invalid retrieval scope. - Retrieval results include a large number of non-compliance documents, with core fields such as compliance status empty. The cause is that
collectionIdsis not specified, and irrelevant knowledge base collections are retrieved, leading results to deviate from the compliance audit scenario. - When retrieving compliance checklists in Word format, the check items corresponding to the table of contents cannot be matched. The cause is that segmentation is not performed according to the table of contents structure of the Word document, so the directory hierarchy cannot be associated with specific compliance items, resulting in failed semantic matching.
How to confirm configurations are correctly set
- On the FastGPT v4.17.0 knowledge base configuration page, check that
collectionIdsonly includes document collections related to cloud resource compliance audit. - Submit a test query such as "Check if current IAM permission configurations are compliant", verify that returned result fields include compliance-related attributes such as resource ID, permission configuration items, and compliance status.
- View the
similaritythreshold in the retrieval configuration, confirm that the value falls within the 0.75–0.85 range. Adjust the threshold and compare retrieval result relevance as needed. - After enabling
usingReRank, verify that reranked results prioritize core compliance documents such as rectification tracking forms or permission audit reports.
The values provided here are common starting points. They should be evaluated against the samples associated with the actual deployment.
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.