What this category of data looks like
Data for this category primarily comes from on-site resources such as customer data center environments, device inventories, software packages, network topologies, and similar on-site assets. Daily generated documents include deployment manuals, installation records, configuration lists, on-site check-in sheets, and other related materials. Data update frequency follows per order or per phase, with updates synchronized alongside the progress of on-site deployment. Data fields include device model, software version, configuration parameters, check-in time, and similar items. There is no unified standardized format, and variations exist based on differences in customer on-site environments.
Constraints Imposed on Knowledge Base Retrieval and Recall by These Characteristics
Data for this category is scattered and updated per order. This requires the knowledge base retrieval and recall process to support fast import and incremental synchronization of multi-source heterogeneous data, to avoid delaying on-site deployment progress caused by full reindexing. Document structures are inconsistent, including multiple formats such as deployment manuals and configuration lists. This requires adapting to contextually relevant recall for different fragments, to ensure retrieved results match actual on-site scenarios. Data varies based on differences in customer on-site environments. This requires retaining sufficient recall entries to cover potential configuration parameters, device information, and similar content, to avoid missing critical details. On-site implementation typically has time sensitivity requirements, so retrieval response speed must align with on-site troubleshooting rhythms, with no overly long wait times.
How to Configure Settings
This document is written based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale for This Setting |
|---|---|---|
similarity | 0.75–0.85 | Configuration parameters, device information, and similar content for typical on-site deployments have strong relevance. This range filters low-relevance results while retaining valid matches |
limit | Top 8 entries | Knowledge base content for typical on-site deployments mostly consists of single-phase device and software-related entries. 8 entries cover information needs for most scenarios, while avoiding overly long context |
usingReRank | Enabled | Typical on-site data has multi-source heterogeneous characteristics. Reranking optimizes the sorting logic of recall results, raising the priority of critical information |
datasetSearchUsingExtensionQuery | Enabled | Typical on-site deployment queries are mostly specific configuration-related inquiries. Enabling this setting generates extended queries based on the original question, covering more potential matching content |
searchMode | Semantic search | Queries for typical on-site scenarios are mostly natural language descriptions of configuration or troubleshooting needs. Semantic search is better suited for actual usage scenarios than keyword matching |
Three Common Misconfigurations
- Symptom: An unauthorized access error appears after configuring authentication for the knowledge base search node. Cause: The role for on-site deployment implementation is not bound to the access permission group of the corresponding knowledge base, or the authentication node is not associated with the correct user identity field.
- Symptom: No optional values appear in the variable reference dropdown menu when selecting a knowledge base. Cause: The data source for the knowledge base variable is not configured in the application workflow, or the knowledge base ID is not output to the current search node via a preceding node.
- Symptom: Chinese garbled characters appear after importing a CSV file to the knowledge base, even after adjusting character encoding settings. Cause: The CSV file was not saved in UTF-8 without BOM format, or the correct encoding parsing parameter was not specified during import.
How to Confirm Successful Configuration
- Run a single simulated retrieval, and verify that the number of returned results matches the configured
limitvalue. - Check the authentication node configuration, and confirm that the current test account can normally access the bound knowledge base content.
- Import a test segment of a deployment manual, retrieve relevant keywords, and confirm that the recall results include the corresponding content.
- Compare the returned results from online conversations and API calls, and confirm that both use consistent retrieval logic and configuration parameters.
The values provided here are common starting points, and should be evaluated 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.