What data looks like for this category
Typically, data for this category comes primarily from art asset libraries, project management systems, engine resource packages, and outsourced delivery files. Daily produced documents usually include art requirement sheets, asset specification descriptions, outsourced feedback forms, and acceptance records. Fields often cover resource numbers, delivery formats, version numbers, acceptance statuses, and similar details. Units involve pixels, polygon counts, file sizes, and other metrics. Data updates follow a daily or weekly cadence. Some urgent outsourced delivery files are updated temporarily alongside project milestones.
What constraints these characteristics impose on knowledge base retrieval and recall
Documents related to art assets often contain specialized parameters such as polygon counts and resolution. Retrieval must match precise fields, and prioritize recalling entries that include specified resource numbers and version numbers. Data from multiple sources has mixed formats including PDF, engine-exported files, and spreadsheets. The system must support text and metadata extraction across different formats. The daily and weekly update cadence requires that incremental recall trigger logic adapts to temporary updates. Large-volume outsourced delivery packages increase text parsing load during retrieval. The volume of resources parsed per batch must be limited. Consistent asset version number fields require that retrieval links version dimensions, to avoid recalling expired assets.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches specialized art asset parameters such as polygon counts and resolution. A higher similarity threshold avoids mistakenly recalling non-matching resources |
limit | Top 10 entries | Limits art asset retrieval results to avoid excessive redundant information interfering with quick confirmation by art-related roles |
searchMode | Exact match | Adapts to fielded content in asset specification documents, and prioritizes recalling entries that include specified resource numbers and version numbers |
usingReRank | Enabled | Re-ranks initial recall results by relevance to specialized parameters, to raise the display priority of core asset information |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonyms related to assets, such as "polygon count" and "triangle face count", to cover terminology habits across different roles |
collectionIds | Bind exclusive collections per project | Isolates art asset data across different projects, to avoid confusion between cross-project resources |
This configuration is based on FastGPT v4.17.0.
Three common configuration mistakes
- Images within art documents in retrieval results cannot be displayed properly. The cause is that the knowledge base only extracts text metadata, and does not configure index paths for associated image resources. This prevents loading original resource links during recall.
- The call order of RAG and MCP tools in workflows is disorganized. The cause is that node dependencies are not configured according to the logic of first retrieving the knowledge base to obtain asset specifications, then calling tools to verify import compliance.
- Knowledge base collection tag entrances cannot be seen in the interface. The cause is that the current account is not assigned management permissions for the corresponding collection, or project-level tag configuration permissions are not enabled.
How to confirm configurations are set correctly
- Manually enter a specialized parameter from an asset specification document, such as "10000 polygon count". Retrieval results must include entries with this parameter, and rank highly.
- Upload a PDF of an art requirement sheet that includes images, to verify that image resources within the document can be loaded normally in knowledge base recall results.
- After binding an exclusive project collection with
collectionIds, retrieve asset documents from other projects. No matching entries should appear, confirming that data isolation is active. - After enabling
usingReRank, compare the order of results before and after re-ranking, to confirm that entries matching core parameters are displayed first.
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.