What data for this category looks like
Typically, knowledge base data for object storage and backup services originates from object storage consoles, backup systems, access logs, and customer policy tables. Daily documents include four categories: bucket configurations, backup policies, recovery records, and lifecycle rules. Fields include bucket names, backup task IDs, recovery timestamps, permission policy content, and other related fields. Data updates occur either via daily synchronization or real-time updates triggered by policy changes. Single document lengths mostly range from 500 to 2000 characters.
What constraints these characteristics impose on knowledge base retrieval and recall
This type of data has strong structured traits and contains permission-sensitive fields. Knowledge base retrieval must first filter unauthorized bucket and policy data. Data updates rely on console operations or policy changes, so a change-triggered synchronization mechanism must be configured to avoid resource consumption from full synchronization. Most documents are configuration entries, so retrieval keywords must match precise fields such as configuration item names and task IDs to prevent irrelevant results from generalized retrieval. Permission policy verification is the most time-consuming step, so retrieval must first associate permission fields for filtering to reduce subsequent verification overhead.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Balance matching accuracy and recall range for structured documents such as bucket configurations and backup policies |
limit | 10–15 items | Adapt to the scale of document entries for object storage and backup services, avoid returning redundant results |
usingReRank | Enabled | Perform relevance reranking on time-series documents such as backup tasks and recovery records, improve result sorting rationality |
datasetSearchUsingExtensionQuery | Enabled | Supplement matching precise fields such as bucket names and backup task IDs, improve retrieval accuracy |
collectionIds | Specify target knowledge base collection ID | Ensure uploaded documents are only synchronized to the knowledge base collection corresponding to the backup service, avoid cross-collection interference |
searchMode | Semantic Matching | Adapt to user scenarios of fuzzy queries, match content such as configuration item names or policy descriptions |
Three common mistakes
- Phenomenon: No new content appears in the knowledge base after calling the upload API. Cause: The
collectionIdsparameter is not correctly specified for the backup service's knowledge base collection, or the bucket name is not carried as an extended query field. - Phenomenon: Retrieval response times out beyond 30 seconds. Cause: The reasonable value of the
limitparameter is not restricted, or bothusingReRankanddatasetSearchUsingExtensionQueryare enabled without adjusting thesimilaritythreshold. - Phenomenon: Retrieval results include unauthorized bucket configurations. Cause: Permission fields from access logs are not associated during retrieval, or policy table data from non-target customers is not filtered.
How to confirm configurations are set correctly
- Call the upload API, then view the document list of the corresponding knowledge base collection in the FastGPT v4.17.0 console, confirm that bucket configuration or backup policy documents have been added.
- Initiate a test retrieval, input "backup policy", check if the number of returned results matches the
limitparameter's setting. - After enabling
usingReRank, compare result sorting before and after enabling, confirm the reranking logic takes effect. - Monitor the retrieval API's response time, confirm it does not exceed 10 seconds.
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.