What the data for this category looks like
Daily data for internal network and WiFi operations comes primarily from switch consoles, wireless controllers, monitoring systems, and user trouble reporting channels, with updates occurring daily or after each network change. Common documents include network topology diagrams, configuration change tickets, fault records, and wireless coverage inspection reports. Data fields include port numbers, VLAN IDs, SSID names, inspection times, fault locations, and more. Some documents use structured table formats, while others are semi-structured fault description texts.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Data sources for this category are scattered and update frequency is high. This requires knowledge base retrieval to balance structured field matching and semantic recall of semi-text content, to avoid missing key change records and fault location information. Questions in operations and maintenance scenarios are mostly for real-time fault troubleshooting, so latest documents from the past 7 days must be prioritized for recall. At the same time, result ranges must be strictly limited to prevent irrelevant other business configuration content from interfering with troubleshooting. In addition, the data includes professional operations and maintenance fields such as ports and VLANs, so retrieval must support precise field matching to avoid excessive invalid information caused by generalized recall.
How to configure the settings
This configuration applies to FastGPT v4.17.0 and later versions.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Fault troubleshooting for internal network operations requires precise matching of core fields to avoid interference from irrelevant historical configuration entries in retrieval results |
limit | Top 8–12 entries | Operations and maintenance personnel need to quickly review core fault records and change tickets; too many entries increase screening costs |
searchMode | Hybrid vector + keyword retrieval | Data includes structured port/VLAN fields and semi-structured fault descriptions; hybrid retrieval balances accuracy and recall completeness |
usingReRank | Enabled | Recall results must be sorted by update time and fault urgency, prioritizing displaying latest change tickets and records related to current faults from the past 7 days |
datasetSearchUsingExtensionQuery | Enabled | Operations and maintenance personnel often omit technical terms when asking questions; extended queries can cover more relevant documents, for example matching "SSID coverage issue" with "wifi is down" |
collectionIds | Bound to operations-specific knowledge base collections | Isolate network operations data from other business data to prevent recall of irrelevant content from affecting troubleshooting efficiency |
Three common configuration mistakes
- Symptom: Garbled characters appear in retrieval results after uploading a CSV format inspection report saved via WPS. Cause: The CSV file encoding was not converted to UTF-8, causing character recognition errors during knowledge base parsing.
- Symptom: The number of returned results during retrieval does not match the set
limit. For example, setting top 10 entries but getting 20 entries returned. Cause: Weight rules for hybrid retrieval were not configured correctly, resulting in duplicate overlapping results from vector retrieval and keyword retrieval. - Symptom: A 504 timeout error is returned after initiating retrieval. Troubleshooting steps: Check if the App/Admin server has configured
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKEN, check the connectivity of the AI Proxy service, and check upstream model channel configurations.
How to confirm the configuration is complete
- Check the environment variables of the App/Admin server, confirm that
AIPROXY_API_ENDPOINTis the root address of the AI Proxy service, formatted similarly tohttp://aiproxy:3000(without appending/v1or/chat/completionspaths). Confirm thatAIPROXY_API_TOKENis an administrator token consistent with the AI Proxy server configuration and non-empty. This variable must not be filled in workflow global variables, application API request headers, or the general model management page. - Initiate a test retrieval, enter a common operations and maintenance question such as "VLAN configuration change records", and verify that the number of returned results, field completeness, and expectations match.
- Confirm that
collectionIdsis bound to the network operations-specific knowledge base collection to avoid recalling irrelevant business data. - View retrieval logs to confirm that no authentication failure or timeout related errors appear in the request.
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.