What the data for this category looks like
This category’s data comes primarily from VPC planning documents exported from cloud platform consoles, security group rule configurations, load balancing configuration files, network topology diagrams, and operation and maintenance change tickets. Data updates trigger when a single configuration adjustment occurs. A new version or entry generates after each change. Document structures center on structured configuration items, including fields like network segment range, port mapping, health check threshold, backend server group ID, and more. Most fields have clear units, such as Class C network segments, seconds, percentages, and others.
What constraints these characteristics impose on knowledge base retrieval and recall
Data update frequency triggers in real time with operation and maintenance changes. Retrieval must prioritize matching the latest configuration entries to avoid returning outdated rules. Structured configuration fields require retrieval to support precise matching of field names and values, plus fuzzy matching of descriptive text. Single configuration documents have long content and link to multiple parameter groups. Limit the number of recalled results to avoid exceeding the context window. Data is divided by project or change ticket. Retrieval must support limiting search scope to exclude unrelated configuration content.
How to set configurations
| Configuration Item | Recommended Value | Rationale for This Setting |
|---|---|---|
similarity | 0.75–0.85 | Configuration items for this category are mostly standardized numerical values. This interval balances precise matching and recall coverage |
limit | Top 8–12 entries | Single load balancing configuration documents have long content. Excessive recall causes context overflow and affects model generation results |
searchMode | semantic + keyword | This category has both structured configuration fields and descriptive text. A mixed mode balances precise matching and semantic association |
usingReRank | Enabled | This category has many configuration items with similar names. Reranking improves the display priority of highly relevant results |
datasetSearchUsingExtensionQuery | Enabled | Configurations for this category often link to auxiliary documents like change tickets and topology diagrams. Extended queries cover associated data |
collectionIds | Specify the collection ID of the target project | Data for this category is divided by project or change ticket. Limiting the collection excludes unrelated configuration content |
Three common mistakes
- The symptom is a configuration error prompt popping up after clicking the knowledge base new button. Troubleshooting steps include verifying incorrect configuration of the
AIPROXY_API_ENDPOINT(root address of the AI Proxy service, do not append paths such as /v1 or /chat/completions, not a single model service address) andAIPROXY_API_TOKEN(AI Proxy administrator Token, must match server configuration and be non-empty, not an application API Key or model vendor key, etc.) environment variables on the App/Admin server. Missing this configuration triggers startup verification failure. - The symptom is search results including load balancing configuration content from non-target projects. The cause is failure to limit the target knowledge base collection scope via the
collectionIdsparameter. - The symptom is image paths failing to load correctly in uploaded MD format documents. The cause is failure to configure knowledge base resource path mapping rules, making image resources unassociable during retrieval.
How to confirm configurations are set correctly
- Check environment variable configurations on the App/Admin server. Confirm that
AIPROXY_API_ENDPOINT(root address of the AI Proxy service, do not append paths such as /v1 or /chat/completions, not a single model service address) andAIPROXY_API_TOKEN(AI Proxy administrator Token, must match server configuration and be non-empty, not an application API Key or model vendor key, etc.) are correctly filled and non-empty. This requirement applies to FastGPT v4.17.0 and later versions. - Run a retrieval test for load balancing configurations. Verify that the number of returned results matches the range set by the
limitparameter. - Test passing the
collectionIdsparameter. Confirm that retrieval results only include configuration content from the specified collection. - After enabling
usingReRankanddatasetSearchUsingExtensionQuery, compare the correlation changes of retrieval results.
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.