What the data for this category looks like
Ledger data for pharmaceutical preparations manufacturing comes from approved formulation processes, material batch information, electronic batch records from MES systems, equipment and environmental monitoring data, test results from LIMS systems, and deviation and change records from quality management systems. Data is updated per production batch, each key process step execution, and per fixed monitoring plans. Daily documents include master production files, batch production records, batch packaging records, cleaning records, test records, deviation/CAPA/change records, and batch release review forms. Fields include material batch numbers, process execution parameters, environmental monitoring values, test results, deviation descriptions, approval status, and more. There are no universal standard units, and industry standard units for corresponding business fields must be used.
What constraints these characteristics impose on the tool calling and plugins workflow
Data for this category is scattered across multiple heterogeneous systems. Tool calling must support connecting to multiple types of data sources including MES, LIMS, and quality management systems, to avoid incomplete ledger summaries caused by data silos. Data updates are tied to production batches and process steps, so tool trigger timing must match production nodes to ensure obtained ledger data aligns with current production status. Document fields have no unified format, so precise extraction of specified business fields is required, and tool calling must support custom field mapping rules. GMP requirements mandate complete and traceable data, so tool return results must retain original data sources and operation logs to facilitate subsequent audits. An exception capture mechanism must also be built in to handle missing ledger data or erroneous reports caused by system call failures.
How to set up the configuration
This document is written for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Adapt to multi-data-source ledger summary requests, automatically determine whether to trigger tool calls |
| Workflow/API function-call configuration | Triggered by batch record fields | Bind key nodes of production batches to ensure call timing matches data update rhythm |
| Tool call timeout | 300-600 seconds | Single-batch ledger involves pulling data from multiple systems, reserve sufficient response time |
| Tool call retry count | 1-3 times | Respond to temporary system connection fluctuations, avoid summary interruption caused by single call failure |
| Retry interval | 10 seconds | Reasonable duration to wait for system connection recovery |
| Maximum number of returned entries per query | Top 100 entries | Avoid pulling excessive redundant data that affects summary efficiency |
| Result output mode | Return only tool call data | Meet the requirement of not outputting AI replies at the tool call node |
| Field mapping rules | Match preset fields per GMP documents | Ensure extracted ledger fields comply with audit requirements, avoid data misalignment |
Three common errors
- Issue: Connection failure or permission denied error when calling database plugins. Cause: The database access whitelist for the environment where FastGPT is deployed has not been configured, or plugin connection parameters do not match the database instance's port and authentication method.
- Issue: The tool call node outputs both AI replies and tool return data simultaneously. Cause: The result output mode is not set to return only tool call data, or the configuration was not saved and activated correctly.
- Issue: Tool calls cannot be triggered normally when calling external models. Cause: The
toolChoiceparameter is not set toauto, or the model does not support standard function call formats.
How to confirm the configuration is complete
- View FastGPT tool call logs to confirm that requests to all associated systems have been successfully initiated and returned valid business data.
- Trigger a ledger summary request for a simulated production batch, check that output content only includes field data returned by the tool, with no additional AI-generated content.
- Adjust test fields in the field mapping rules to verify that extracted ledger data fully matches preset business fields.
- Simulate a system connection failure scenario to confirm that the tool call triggers the preset retry mechanism and completes automatic recovery.
The values provided here are common starting points and should be evaluated against the reader's own samples.