What this category of data looks like
Petroleum refining scenario data sources include DCS/APC systems, LIMS laboratory information management systems, tank farm metering systems, crude oil evaluation databases, equipment management systems, and energy management systems. Data update frequencies include real-time, shift-based, and daily. Daily documents include operating procedures, production plans, material balance sheets, distillate export quality reports, interlock ledgers, shift handover records, and change management documents. Data fields include equipment numbers, inspection points, operating temperature, pressure values, maintenance man-hours, and more. Units include MPa, ℃, cubic meters, hours, and others.
What constraints these characteristics impose on the "tool calling and plugins" workflow
Multi-source data updated in real time and at scheduled intervals requires tool calling to support low-latency responses and concurrent pulling of multiple interfaces, to avoid delaying inspection and maintenance decisions due to data lag. Diverse document formats require plugins to adapt to text parsing for different types such as structured reports, ledger records, and process procedures. Dynamic fluctuations in raw material properties and unit load require tool calling to associate crude oil evaluation and DCS operating data in real time, to assist in judging equipment risks. Severe consequences of unplanned shutdowns and other errors require plugins to be configured with failure retry mechanisms and abnormal status reporting logic.
How to set the configuration
This document is compatible with FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Adapts to the need for automatic switching between tool calling and rule-based responses in equipment inspection Q&A |
| Workflow/API function-call configuration | required | Forces generation of calling requests that conform to equipment maintenance parameter formats, to avoid format errors |
| Tool call timeout | 60 seconds | Matches the update rhythm of real-time data from refining equipment, to avoid delayed decisions caused by timeouts |
| Document parsing batch size | 800–1200 characters | Adapts to the typical text length of refining operating procedures and maintenance records |
| Multi-tool concurrency limit | First 3 | Matches the parallel data pulling requirement of multiple continuous units coupled through logistics and hydrogen energy networks, to avoid overloading system resources |
| Abnormal retry count | 2 times | Addresses network fluctuations or temporary interface failures, to reduce the probability of single call failure |
Three common mistakes
- Phenomenon: When calling the embedding model,
{"object":"error","message":"Only allowed now"}is returned. Cause: The whitelist permission for model calling is not configured, or the corresponding model is not deployed and bound in the FastGPT backend. - Phenomenon: After configuring HTTP tool calling, the interface request log shows normal parameter parsing, but the target device has no execution feedback. Cause: The request execution permission is not enabled in the plugin configuration, or the interface address is not added to the system's allowed outbound domain name list.
- Phenomenon: After uploading equipment inspection images, the base64-encoded interface request returns missing parameters. Cause: The base64 encoding prefix identifier is not configured correctly, or the encoded image data is not carried correctly in the request body.
How to confirm the configuration is complete
- Initiate a query request for inspection parameters of a specific refining equipment, and check whether the interface address and incoming parameters in the tool call log match the configuration.
- Simulate interface timeout or error responses, to verify whether the system triggers the configured retry logic and abnormal recording.
- Import a distillate export quality report document, and check whether the plugin can extract corresponding fields according to the configured parsing rules.
- Switch the Q&A scenario, to verify whether the
functionCallconfiguration can force the generation of tool calling requests that conform to the format.
The values provided are common starting points and should be measured against the reader's own samples.