What this category of data looks like
Data for this category typically comes from internal enterprise procurement systems, supplier quotation documents, budget spreadsheets, and contract systems. Core daily generated documents include procurement requisitions, price comparison records, formal contracts, and acceptance documents. Data updates are tied to individual procurement workflows: full data updates are completed for a single procurement from application to acceptance, while regular monthly bulk procurements sync full supplier and budget data on a monthly basis. Document fields include application number, procurement category, budget amount, shortlist of candidate suppliers, contract clause number, default agreement, and more. There is no unified fixed unit field; the pricing unit matches the associated procurement category.
What constraints these characteristics impose on the tool calling and plugins workflow
Data sources for this category are scattered and have significant format differences. The tool calling link must adapt to access requirements for multi-system data sources, and corresponding plugins must be configured to pull and parse data from different systems such as procurement and contract systems. Data updates are triggered primarily by individual procurement workflows, so tool calling must be bound to specific steps in the procurement workflow to avoid invalid triggers. Field structures vary significantly across different documents, so parameter extraction rules for tool calling must be configured separately for different document types such as procurement requisitions and contracts to ensure accurate field matching. Errors in this link can directly lead to procurement delays or budget overruns, so failure handling for tool calling must be configured with retry mechanisms and exception alert logic to ensure workflow stability.
How to set configurations
The following configuration instructions are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
toolChoice | auto | This category's tool calling scenarios cover multi-data-source pulling and field extraction. Auto mode can adapt to dynamic requirements of different documents |
functionCall | enabled | This category requires calling multi-system plugins to complete data pulling and parsing. Enabling function calling enables accurate parameter passing and result parsing |
| Single tool call timeout | 300 seconds | Interface responses from procurement and contract systems typically complete within 3 minutes. Setting a timeout prevents workflow blocking |
| Tool call result cache duration | 7 days | Historical data for monthly bulk procurements must be retained for at least one settlement cycle. Caching improves efficiency for repeated calls |
| Exception retry count | 2 times | Call failures caused by temporary system fluctuations can be recovered via retries. Excessive retries will extend workflow cycles |
| iframe cross-origin whitelist configuration | Include domains of procurement and contract systems | Tool calling for this category often requires embedding document resources from multiple systems. Configuring the whitelist ensures normal resource loading |
Three common misconfigurations
- Symptom: Embedded supplier quotation or contract documents via iframe fail to display images and user avatars. Most browsers fail to load, but only Quark browser works normally. Cause: The iframe cross-origin whitelist is not configured. Some browsers' cross-origin policies restrict resource loading from untrusted domains.
- Symptom: In a two-layer tool calling workflow, the AI return content from the first layer of tool calling is printed to the workflow log and cannot be hidden. Cause: The intermediate step hiding setting for the workflow is not enabled. By default, execution content for all nodes is output.
- Symptom: The tool calling module outputs both AI reply content and triggers the next-level branch execution, making it impossible to execute only one of the two. Cause: Dependent conditions for branch triggering are not set correctly, resulting in parallel execution of multiple workflow nodes.
How to confirm correct configuration
- Initiate a simulated procurement requisition workflow to trigger the tool calling link, and verify that historical budget data and supplier quotation information from the procurement system are successfully pulled.
- View the workflow execution log and confirm that only necessary tool call results are output, with no redundant AI reply content or intermediate step logs.
- Simulate a temporary system timeout scenario and confirm that the tool calling triggers the preset retry mechanism and exception alert.
- Test the loading of iframe-embedded contract documents across different browsers and confirm that cross-origin resources display normally.
The values given are common starting points and should be measured against the reader'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.