Tool Calling and Plugins for On-Call Transfer and WeChat

Data Characteristics for This Category

On-call transfer scenarios involve data centered on staff schedules, contact information, transfer rules, and historical transfer records. Data sources typically include internal scheduling systems, HR systems, or specialized on-call management platforms. This data updates frequently, potentially daily or weekly. The document structure is primarily structured data, often in JSON or database table records. Key fields include shift_id (shift ID), on_duty_staff_id (on-duty staff ID), start_time (start time), end_time (end time), contact_info (contact details, such as WeChat Work ID or phone number), and escalation_policy (escalation policy). Time fields usually follow ISO 8601 format. The contact_info field may contain various types, requiring parsing.

Constraints Imposed by These Features on Tool Calling and Plugins

On-call transfer data characteristics impose specific requirements on tool calling and plugins. First, schedule information is time-sensitive. Tool calls must retrieve the latest data in real time to prevent transfer failures due to outdated information. This means caching mechanisms need careful design, or tools should directly query source systems. Second, diverse contact information requires plugins to handle various contact_info formats. For example, a plugin might parse a WeChat Work ID into usable API parameters or identify a phone number to call a corresponding communication tool. Third, in case of transfer exceptions, the escalation_policy field determines the next automated process. Tools must be able to call different plugins or services based on this policy to implement multi-level transfers or notifications. Accurate matching of data fields and consistent units (e.g., milliseconds or seconds for timestamps) are critical for successful tool calls.

Configuration Guidelines

Configuration Item	Recommended Value	Rationale
`tool_timeout`	`30 seconds`	On-call transfers require quick responses. Avoid long waits that could disrupt service.
`max_retries`	`2 times`	Appropriate retries improve success rates during network fluctuations or temporary service unavailability.
`param_mapping`	`on_duty_staff_id` maps to `user_id`	Ensure FastGPT internal fields align with external scheduling system fields.
`json_schema_validation`	`enabled`	Strictly validate data types and formats of incoming parameters, for example, `contact_info` as a string.
`stream_output`	`disabled`	On-call transfers typically require a single result. This reduces interference from intermediate states in streaming output.
`cache_ttl`	`600 seconds`	Schedule information often updates hourly. Short-term caching improves query efficiency.

Three Common Mistakes

Tool call results show an empty user_id field: This usually indicates incorrect param_mapping configuration. The system failed to correctly map the staff identifier from the scheduling system to the user_id field required by the tool.
On-call transfer notifications are delayed or fail: This may be due to tool_timeout being set too short, causing the tool call to prematurely abort during poor network conditions, or external service response times exceeding expectations.
The tool calling module in the workflow does not trigger: This often happens when workflow logic branch conditions are inaccurate, preventing the system from entering the path containing the tool call, or the output of the preceding module for the tool call does not meet expectations.

How to Confirm Correct Configuration

In the FastGPT debug interface, select a typical on-call transfer scenario. Observe the output logs of the tool calling module. Confirm that key fields such as on_duty_staff_id and contact_info are correctly parsed and passed.
Manually simulate an on-call transfer request. Check if the actual WeChat Work group notification or phone call is accurately sent to the current on-duty staff. Verify that the notification content matches expectations.
Use FastGPT's monitoring features to observe the success rate and average response time of tool calls. Ensure these metrics operate within acceptable business thresholds.

Note: The values provided are common starting points. Measure them against specific samples to determine optimal settings for a particular use case.

Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.