What the data for this category looks like
IT asset inventory data comes from asset management systems, procurement systems, financial fixed asset modules, and terminal management platforms. Daily maintained documents include asset ledgers, inventory sheets, receipt and return forms, and scrap application forms. Documents contain fields such as asset number, name, specification model, user, location, purchase date, and usage status. Units are mostly sets, pieces, or units. Data update frequency is mostly monthly, quarterly, or annually. Document formats and field details vary across different scenarios.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Multi-source data requires high field alignment. Retrieval must match unified asset identifiers exported from different systems to prevent data confusion. Different update frequencies dictate the knowledge base synchronization rhythm, requiring incremental update tasks to align with monthly or quarterly inventory cycles. Diverse document types mean retrieval must cover content from scenarios like ledgers and receipt records, while prioritizing core fields such as asset number and user. Asset data accuracy directly ties to audits and physical-book reconciliation, so recall result precision is higher than in general scenarios.
How to set configurations
The following configurations are adapted for FastGPT v4.17.0, formulated based on the business characteristics of IT asset inventory:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Asset data fields are clearly defined, requiring filtering of low-match irrelevant entries to avoid incorrect associations during physical-book reconciliation |
limit | Top 10 entries | The volume of single knowledge base entries for documents related to asset inventory is manageable; excessive results increase context processing load |
searchMode | vector+keyword hybrid mode | Asset data includes both structured fields and unstructured descriptions; hybrid retrieval covers precise number queries and fuzzy description queries |
usingReRank | Enabled | Recall results must be sorted by relevance to ensure core asset information is returned first, meeting the rapid positioning needs of pre-sales and technical support |
datasetSearchUsingExtensionQuery | Disabled | Queries for asset inventory are mostly clear asset attribute queries, no additional question expansion is needed to avoid introducing irrelevant interference |
Three common mistakes
- An
embedding erroroccurs when uploading asset ledger documents. The cause is that the document contains non-standard asset numbers or special characters, causing the embedding model to fail processing. - The number of returned results when searching asset receipt records does not meet expectations. The cause is that the
limitparameter is not configured correctly, and the value does not match the document volume of asset inventory, leading to too many or too few returned results. - Relevant records cannot be accurately recalled using the asset user field. The cause is that the hybrid retrieval mode of
searchModeis not enabled; relying only on vector retrieval cannot match structured user fields.
How to confirm the configuration is correct
- Upload a standard-format asset ledger document, check that the embedding process completes normally with no error prompts.
- Enter a specific asset number to initiate a query, check that the returned results include complete field information for the corresponding asset, with matching accuracy meeting expectations.
- After enabling
usingReRank, verify that the returned results are sorted from highest to lowest relevance, with core asset information displayed first. - Test retrieving both inventory sheets and receipt and return forms at the same time, confirm that the results cover valid content from the corresponding document types.
The values provided here 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.