What data is included in this category
Data for this category originates from monitoring alerts, logging platforms, APM tools, and user feedback. Daily generated documents include trouble tickets, incident timelines, emergency response plans, and incident notifications. Data updates occur in real time or are synchronized immediately after each incident is triggered. Data fields cover alert trigger time, affected service scope, recovery operation records, ticket number, notification recipients, and no additional aggregated statistical fields. Only the original associated data for a single incident is retained. This guide is developed for FastGPT v4.17.0, and is compatible with the tool calling and plugin configuration logic of this version.
What constraints these characteristics impose on tool calling and plugins
Because data sources are scattered and formats are inconsistent, tool calling must support format adaptation for multi-source data, and avoid hardcoding field rules for a single data source. The real-time update feature requires that the timeout threshold for tool calling matches the incident response window. Excessive waiting will delay stop-loss operations. The single-incident document structure limits the input length of context, so only ticket and timeline data associated with the current incident must be loaded. Additionally, the high priority of troubleshooting requires that a preset downgrade process is triggered when tool calling fails, to avoid further delays in service recovery due to tool calling failures.
How to configure the settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
toolChoice | auto | Adapts to tool calling requirements for different troubleshooting scenarios, allows the system to automatically select appropriate tools |
functionCall | First 3 matching tools | Limits the scope of tool calling, avoids redundant operations that slow down incident response |
| Tool calling timeout | 60 seconds | Matches the response window of SRE on-duty shifts, prevents excessive waiting that delays stop-loss operations |
| Plugin triggering rules | Trigger based on alert severity | Differentiates between high and low priority incidents, only calls tools during high-priority events |
| Context loading limit | 800–1200 characters | Restricts the volume of document data for a single incident, adapts to the length of real-time updated incident data |
| Number of failure retries | 2 | Balances tool calling success rate and response speed, avoids excessive retries that extend troubleshooting time |
Three common misconfigurations
- Symptom: After the variable update tool runs, variables in the incident ticket are not automatically synchronized. Cause: The variable binding rule for
functionCallis not configured, and the tool calling is not associated with the variable update field of the incident. - Symptom: Calling a plugin returns an empty result with no error prompt. Cause: No data source filtering conditions are set for plugin triggering, and irrelevant historical incident data is loaded, causing the context to exceed the limit.
- Symptom: Uploading files via the API call workflow returns a format error. Cause: Log files exceeding the preset size or non-standardized trouble ticket attachments are uploaded, not following the file format requirements for incident response scenarios.
How to confirm the configuration is correct
- Manually trigger a high-priority alert test, check whether the tool automatically calls the preset troubleshooting recovery plugin.
- View the tool calling logs, confirm that the configurations for
toolChoiceandfunctionCallare correctly loaded. - Upload a test trouble ticket file, verify that the API call format complies with the preset rules.
- Simulate a tool calling failure scenario, confirm that the system triggers the preset downgrade process.
The values provided here are common starting points, and should be evaluated against the user'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.