What data for this category looks like
Data for this category primarily comes from cloud management platforms, order systems, IAM permission systems, and billing systems. Daily documents include resource activation forms, instance configuration sheets, account permission forms, and delivery confirmation forms. Each single data entry is associated with one cloud host instance, and includes fields such as instance specifications, deployment region, network configuration, permission binding information, and delivery status. Updates are synchronized in real time for each order, and the data volume for a single delivery process grows linearly with the number of instances.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Since data for this category is updated in real time via delivery orders, the knowledge base must support incremental synchronization of order, permission, and billing data to avoid retrieving expired information. Each single delivery data entry includes strongly related fields such as instance specifications, region, and permissions. During retrieval, keywords such as instance ID or region from user questions must be prioritized to narrow the recall scope. Additionally, specification verification and security group configuration steps in the delivery process rely on frequently reused configuration documents, so these documents must be prioritized for recall to reduce manual verification time. Furthermore, the data structure of different delivery orders is highly uniform, so retrieval can quickly locate relevant documents through fixed fields without performing a generalized search of the full knowledge base.
How to set the configuration
The following settings are based on the knowledge base retrieval module of FastGPT v4.17.0:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for strongly related fields such as instance specifications and region in this category, avoiding recalling irrelevant cloud network configuration documents |
limit | Top 8 entries | The number of associated documents for a single delivery process in this category typically does not exceed 7, so retaining redundancy ensures coverage of all necessary configurations |
searchMode | semantic | Matches keywords such as instance ID and specifications from user questions, and is more suitable for processing natural language pre-sales inquiries than keyword matching |
usingReRank | Enabled | Security group and permission rules in delivery configurations have semantically similar entries, and re-ranking can improve the sorting priority of highly relevant documents |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands instance specifications from user questions into matching configuration document keywords, reducing the probability of missed retrieval |
collectionIds | Bind to cloud host provisioning-specific knowledge base collections | Avoid recalling configuration documents from other categories such as cloud networks, narrowing the retrieval scope |
Three common mistakes
- Phenomenon: Unable to correctly reference instance specification variables during retrieval. Cause: Variable mapping function was not enabled in the retrieval configuration, causing variables in the question to not be correctly replaced with field values from documents.
- Phenomenon: CSV-formatted delivery confirmation sheets cannot be retrieved after import. Cause: CSV was uploaded directly without using the template import function, resulting in incorrect mapping of document fields and failure to generate valid vectorization results.
- Phenomenon: Vectorization results for historical delivery orders cannot be loaded during retrieval. Cause: The vector storage association path for the knowledge base was not correctly configured, causing the system to fail to read generated vectorization data.
How to confirm the configuration is correct
- Upload a test resource activation form document, enter a question containing an instance ID in the test interface, and verify whether the retrieval result includes this document.
- Adjust the
similaritythreshold to 0.7, then adjust it to 0.9, observe changes in the number of retrieval results, and confirm that the threshold setting meets business requirements. - After enabling
usingReRank, enter a question containing security group configuration, and verify whether highly relevant documents are sorted earlier. - Check the knowledge base collection binding status, confirm that only cloud host provisioning-specific datasets are bound, and no collections from other categories are included.
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.