What the data for this category looks like
This category's data mainly comes from enterprise OA systems, organizational structure databases, internal system documents, and user feedback. Daily produced documents include process configuration sheets, approval node tables, form field descriptions, and change records. These contain fields such as approval node ID, permission scope, form field name, and effective time. Data update frequency is triggered by process changes. There are usually 1 to several updates per month. Update content mostly involves node adjustments, permission changes, or form field optimizations.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Data for this category is scattered across multiple systems and documents. Retrieval must cover multiple data sources including OA process configurations and organizational permission rules. This creates requirements for cross-source recall accuracy. Documents such as process configuration sheets and approval node tables have fixed structures but closely related fields. Retrieval must accurately match node IDs and permission scopes to avoid interference from irrelevant results. Data update frequency fluctuates with process changes. The recall link must support incremental synchronization of updated documents to ensure retrieval results match currently effective rules. Additionally, change records must serve as auxiliary retrieval basis to help locate recently adjusted process nodes.
How to configure settings
This document applies to official rules for FastGPT v4.17.0. Before starting configuration, confirm that the App/Admin server has correctly configured AIPROXYAPIENDPOINT and a non-empty AIPROXYAPITOKEN. Do not mix these settings with model channel configurations. The following are specific configuration items and recommendations:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirement for process configuration. Prevents irrelevant approval nodes from appearing if set too low, and misses valid configurations if set too high |
limit | Top 8–12 entries | The number of nodes for office process configurations is usually limited. A small number of results can cover complete configuration logic |
searchMode | vector + keyword | Balances semantic matching of process descriptions and keyword matching of structured information such as node IDs and field names |
usingReRank | Enabled | Sorts multiple retrieved configuration content segments to prioritize returning approval rules most relevant to the current query |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms to cover common synonyms in process configurations, such as "approval node" and "audit link" |
collectionIds | Specify the collection ID of the corresponding process configuration knowledge base | Limits the retrieval scope to avoid mixing knowledge base data from other categories and improve retrieval precision |
Three common mistakes
- Symptom: After uploading a single process configuration document, it displays 8 segments, then changes to 13 segments with repeated segmented content. Cause: No reasonable chunk association parameter is configured, or old version data is not cleaned during knowledge base synchronization, leading to duplicate indexing.
- Symptom: Retrieval results include non-process configuration document content, such as employee account permission descriptions. Cause: The
collectionIdsparameter is not specified. The retrieval scope covers the full knowledge base, leading to irrelevant data being recalled. - Symptom: When initiating an approval node query, returned results contain a large number of irrelevant form field descriptions. Cause: The
similarityparameter value is unreasonable. A value that is too low returns redundant results, and a value that is too high misses matching configuration content.
How to confirm configuration is complete
- Check the knowledge base index log to confirm only target process configuration documents are indexed, with no data from other categories mixed in.
- Run a test query, enter a specific approval node name, and verify that the number of returned result segments matches the original document, with no repeated content.
- Adjust the
similarityparameter range to verify whether the precision of retrieval results meets the query requirements for process configurations. - Trigger an update to a process configuration, confirm that after the knowledge base completes synchronous updates, the new configuration content can be retrieved normally.
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.