What data is included in this category
This covers full-process material data from raw material warehouses, line-side warehouses, finished goods warehouses, and material control departments. Daily documents include receiving slips, putaway slips, picking and return slips, inventory transfer slips, inventory count sheets, lot ledgers, first-in first-out (FIFO) and expiration date records.
Data sources include ERP/WMS systems, barcode or RFID equipment, purchase and work order information, bin location records, AGV or conveyor system data, and weighing data. Update frequencies fall into three categories: real-time (such as bin location changes), daily (such as picking and return statistics), and monthly inventory reconciliation (such as book vs physical inventory checks). Core fields include material code, lot number, bin number, inspection status, expiration date, and inventory quantity.
What constraints do these characteristics impose on source citation and traceability?
Integrate multi-source data by linking ERP purchase information, WMS bin location records, and real-time barcode scan data to ensure the traceability chain fully covers the full process from material procurement to storage. Adapt sync logic for real-time and offline data across sources with different update frequencies to avoid traceability results showing discrepancies between book and physical inventory.
Use unique identifiers such as lot number and material code to precisely match core fields and recall corresponding documents, supporting verification steps for supplier and process quality audits. Support batch recall of warehouse documents across different scenarios, while blocking irrelevant data from interfering with LLM context processing. Additionally, traceability accuracy directly impacts production workflows, so strictly prevent forged citations or missing critical information.
Configuration settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Aligns with average text length of documents such as receiving slips and lot ledgers, preventing truncation of critical inspection status and expiration date information |
| Single search limit | Top 6 entries | Single lot of in-plant logistics materials typically links to fewer than 6 documents; excess entries increase LLM context burden |
responseData | Return original document metadata | Facilitates extraction of core traceability fields including material code, lot number, and bin number, supporting verification needs for supplier and process quality audits |
| Maximum context | 4000–6000 characters | Aligns with total length of spliced warehouse documents, matching FastGPT v4.17.0 context processing limits |
detail | Enabled | Used to retrieve detailed inspection records and book vs physical inventory difference information, meeting detail requirements for quality audits |
Three common errors
- Phenomenon: When dynamically specifying a knowledge base, the reference variable dropdown menu for the knowledge base search node is blank with no options. Cause: The knowledge base bound to the ERP/WMS data source is not configured as a reference variable source, or the variable is not set to a type allowed to be called by the knowledge base search node.
- Phenomenon: Warehouse data from different sources cannot be distinguished in logs. Cause: Data source types (such as receiving slips, inventory transfer slips) were not marked when uploading documents, or documents from different data sources were not stored in separate libraries for differentiation.
- Phenomenon: LLM output includes the phrase "Citation: [1]". Cause: Explicit rendering configuration for citation markers was not disabled, or output logic retaining original citation identifiers was not adjusted.
How to confirm correct configuration
- Trigger a test query, enter a question containing a specific lot number, and verify that returned results include warehouse document metadata for that lot.
- View the running logs of the knowledge base search node, confirm that the data source tags of returned documents match the actual warehouse data source types.
- Check the LLM output content, confirm that unapproved citation marker phrases do not appear.
- Adjust the value range of the
quoteMaxTokenparameter, verify that long-text warehouse documents are fully recalled without truncating core fields.
The values given are common starting points and should be measured against the reader's own samples.