What data for this category looks like
Data for this category comes primarily from on-site first-hand resources including customer data center environments, equipment lists, software packages, and network topologies. Daily outputs include deployment manuals, installation records, configuration lists, and on-site check-in forms. Update frequency is once per order or each implementation phase. Each individual data set includes fields such as equipment model, IP address, software version, configuration parameters, and on-site check-in information. There are no unified fixed units, and values are adjusted based on actual on-site environment conditions.
What constraints do these characteristics impose on knowledge base retrieval and recall
This category of data is mostly temporary resources for single projects, with an update frequency of once per order or each implementation phase. Knowledge base retrieval must quickly adapt to newly added single-project data. Documents include multiple types such as deployment steps, configuration parameters, and on-site records. Fields have no unified fixed format, so the recall link must support text content with different structures. Data comes from customer on-site environments and may contain sensitive information. Retrieval scope must be limited to datasets associated with the current project to prevent cross-project data leaks. The size of individual data sets fluctuates with project complexity, so retrieval response requirements for different data volumes must be supported.
How to configure settings
The following configurations are optimized for on-site deployment implementation scenarios based on FastGPT v4.17.0:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches configuration parameter text for on-site deployments. Avoids introducing irrelevant results if set too low, and avoids missing valid configuration items if set too high |
limit | Top 3–5 results | On-site deployment documents are mostly dedicated to single projects. A small number of results can cover on-site troubleshooting needs, and avoids interference from excessive information |
usingReRank | Enabled | Step-by-step text for on-site deployments has similar phrasing. Re-ranking improves the sorting priority of key steps |
datasetSearchUsingExtensionQuery | Enabled | Supplements customer on-site environment descriptions as extended queries to improve recall accuracy for specific on-site equipment |
collectionIds | Bind exclusive dataset ID for the current project | Limits retrieval scope to deployment documents for the current project to prevent cross-project data leaks |
Three common configuration errors
- Slow indexing speed after calling the knowledge base interface to create a file collection. The cause is failing to limit the scope of the current project's
collectionIds, so retrieval scans the full dataset and increases retrieval overhead. upstream connect error or disconnect/reset before headerror occurs when uploading files. The cause is failing to configure the local deployment network policy for the current project's dataset, resulting in failure to connect to the local vector database service during interface calls.- After Excel import, auto-split functionality merges multiple rows into a single entry. The cause is failing to correctly set custom delimiters, failing to split each row as an independent question-and-answer entry, and failing to match the single-line record format of on-site deployment documents.
How to confirm configuration is complete
- Call the knowledge base interface with the current project's
collectionIdsparameter, and check that returned results only include deployment documents, installation records, and other exclusive data for the current project. - Enter the IP address or software version of an on-site device as a query term, and check that the recall results include the corresponding configuration list content.
- Upload an Excel file and set custom delimiters, and check that each imported record corresponds to one row of the Excel sheet.
- Trigger retrieval and view the re-ranked result order, and confirm that key deployment steps appear at the top.
The values provided 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.