What the data for this category looks like
This category’s data comes from server configurations, middleware configurations, security device policies, and baseline tool output results. Daily associated documents include baseline checklists, configuration verification records, rectification suggestions, and exception approval forms. Fields covered include configuration item identifiers, compliance check results, rectification requirements, and exception approval statuses. The update frequency for this data is monthly or quarterly. Each individual document contains a large number of independent configuration items. Configuration items differ across different business environments.
What constraints these characteristics impose on knowledge base retrieval and recall
This category’s data is multi-source and heterogeneous, and each document contains a large number of independent configuration items. This leads to context redundancy during retrieval. The monthly or quarterly update cycle requires the knowledge base retrieval link to support periodic incremental synchronization. This avoids resource consumption from full re-scans. The existence of non-standard documents such as exception approval forms requires retrieval to support both exact matching of configuration items and fuzzy matching of exception reasons. Differences in configuration items across environments require that retrieval scope can be divided by business scenario. This prevents invalid cross-environment matching.
How to Configure Settings
The following configurations are adjusted for FastGPT v4.17.0:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets exact matching requirements for security baseline configuration items, filters low-correlation configuration check results |
limit | Top 8–12 items | Controls the number of configuration items recalled per query, avoids context overload, and aligns with operations staff’s habit of quickly locating issues |
searchMode | Hybrid vector search + keyword search | Balances exact keyword matching for configuration items and semantic vector recall, adapting to multi-source heterogeneous baseline data |
usingReRank | Enabled | Re-sorts recall results to elevate the display priority of high-match configuration items, reducing manual screening effort |
datasetSearchUsingExtensionQuery | Enabled | Automatically generates extended query terms based on user questions, covering configuration item aliases and industry-standard expressions, adapting to colloquial queries from operations staff |
collectionIds | Configured by business environment grouping | Isolates baseline knowledge bases for different server and middleware environments, preventing invalid cross-environment recall |
Three Common Configuration Errors
- Generated output includes the phrase "Reference Marker: [1]". This occurs when the reference annotation toggle for knowledge base retrieval results is not disabled, causing document identifiers from matched retrieval results to be directly appended to generated content.
- Retrieval result granularity does not meet business requirements. This occurs when reasonable chunking rules are not set for individual configuration items in baseline checklists, leading to recall of overly redundant large content or over-splitting of valid configuration items.
- Retrieval results do not match the current operations environment. This occurs when the
collectionIdsparameter is not configured by business environment, causing incorrect recall of cross-environment baseline data.
How to Verify Successful Configuration
- Submit a query that includes a specific configuration item name, and check whether the match quality of recall results meets expectations.
- Review the order of retrieval results to confirm that the re-ranking function is active, and that high-match configuration items are displayed first.
- Switch between different business environment groups, and confirm that only baseline knowledge base data from the current environment is recalled.
- Check generated content to confirm that no redundant reference marker-style text appears.
The values given 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.