What the data for this category looks like
The data for this category primarily comes from CRM systems, contract order databases, identity management systems, and product backends. Daily generated documents include onboarding application forms, tenant configuration sheets, administrator account lists, and delivery confirmation documents. Core fields include tenant unique identifiers, customer entity names, administrator account information, permission configuration items, and onboarding statuses. Data updates occur per order or once daily, synchronized with the progress of customer onboarding workflows. This category differs from customer data import and migration services. Tenant account and basic configuration creation takes priority.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Data sources for this category are scattered but closely linked. Knowledge base retrieval must support cross-system field matching to avoid configuration errors caused by information silos. Data updates proceed in real time alongside customer onboarding workflows. Retrieval must adapt to frequently updated documents to ensure recalled content is the latest tenant configuration and permission information. Daily document fields are fixed and responsibilities are clearly defined. Retrieval must accurately match core fields such as tenant ID and administrator account to reduce invalid recall results. Additionally, the most time-consuming step for this category is customer information verification, and errors will directly impact customer login and delivery acceptance. The accuracy and timeliness of retrieval and recall must be prioritized.
How to set the configurations
This configuration guide applies to FastGPT v4.17.0
| Configuration Item | Recommended Value | Rationale for This Value |
|---|---|---|
similarity | 0.75–0.85 | Meets the precision requirements for matching tenant configuration documents, while covering basic recall scope |
limit | Top 6 entries | Single documents for this category have moderate information volume. 6 entries can cover all associated configurations for a single customer onboarding workflow |
searchMode | Exact match | Matches fixed fields such as tenant ID and administrator account, to avoid irrelevant content introduced by fuzzy retrieval |
usingReRank | Enabled | Sorts recall results by relevance, to prioritize documents most relevant to the current customer onboarding request |
datasetSearchUsingExtensionQuery | Disabled | Document structures for this category are fixed, no additional query term expansion is needed, to avoid introducing irrelevant information |
collectionIds | Bind exclusive tenant configuration knowledge base collections | Isolates data from this category and other business knowledge bases, to prevent cross-category retrieval interference |
Three common mistakes
- Scanned format e-books cannot extract valid text content after upload. The cause is that document OCR recognition configuration is not enabled. The system parses only native text format documents by default.
- Error prompts pop up continuously when adding a knowledge base. The cause is that the storage permission for the knowledge base collection is not configured correctly, or the uploaded document format does not meet system support specifications.
- Retrieval results are abnormal when the user’s query is overly long. The cause is that query term truncation rules are not configured. Excessively long input will cause vector generation deviation, which affects recall accuracy.
How to confirm configurations are set correctly
- Upload a test tenant configuration document, confirm that the returned results include the core field content of this document after executing retrieval.
- Adjust the
similarityparameter value, verify that the relevance of retrieval results changes as expected with the parameter adjustment. - Call the retrieval interface with the tenant ID as the query term, confirm that the returned results only associate configuration documents for this tenant.
- After enabling the
usingReRankconfiguration, verify that the sorting order of recall results follows relevance priority.
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.