What the data for this category looks like
Data for this category originates 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 inventories, and rectification tracking forms. Data updates occur monthly or quarterly. Document fields include resource unique identifiers, permission configuration items, compliance check statuses, rectification tracking progress, and other fields. No unified fixed units apply. Some fields rely on the native identification systems of corresponding cloud vendors.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Data sources are scattered and formats are heterogeneous. The retrieval link must support adaptation and unified parsing of multi-source formats. Update frequency is low, but single data volume is large. Balance must be struck between retrieval efficiency and full coverage. Documents focus on three core topics: permissions, configurations, and evidence retention. Retrieval must accurately match compliance-related keywords. Rectification tracking forms have timeliness requirements. Retrieval results must associate the latest compliance status to avoid using outdated data. The resource inventory counting process is time-consuming. The recall link must reduce redundant calculations to avoid prolonging overall processing duration.
How to configure settings
The following configurations are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches keyword density and field relevance of cloud resource compliance documents, filters irrelevant non-compliance data |
limit | Top 8–12 entries | Covers the three core document types required for compliance audits: permissions, configurations, and rectification, while controlling the context length of a single retrieval |
searchMode | vector | Prioritizes matching semantically related compliance checklists and permission audit reports, adapts to multi-source heterogeneous compliance document structures |
usingReRank | Enabled | Rearranges recall results by compliance relevance, optimizes result sorting for queries about permission anomaly identification |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands compliance-related keywords based on user queries, improves recall accuracy for resource inventory counting queries |
collectionIds | Bind exclusive compliance document collections | Isolates compliance audit data from other category data, prevents non-compliance documents from appearing in retrieval results |
Three Common Misconfigurations
- Symptom: An
upstream connect error or disconnect/reset before headerror is returned when uploading compliance documents to the knowledge base. Cause: The exclusive compliance document collection is not bound, causing the retrieval request to trigger cross-collection upstream connection timeout. - Symptom: QA split collections created via the OpenAPI knowledge base interface have slow retrieval speed and poor effect. Cause: The
datasetSearchUsingExtensionQueryconfiguration is not enabled. Manual query expansion adds additional overhead. - Symptom: Retrieval results only return a small number of permission configuration documents, and cannot cover rectification tracking form data. Cause: The
limitparameter value is too low, failing to cover all compliance document types.
How to Confirm Configurations Are Correctly Applied
- Run a query related to resource inventory counting, verify that returned results include the three document types: compliance checklists, permission audit reports, and rectification tracking forms.
- View retrieval logs, confirm that values for configuration items such as
similarityandlimitmatch preset configurations. - Upload a new compliance document, confirm no
upstream connect errortype errors are reported during upload. - Compare retrieval result sorting before and after enabling
usingReRank, confirm the reranking logic functions as expected.
The values provided are common starting points and should be evaluated against the user’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.