This guide is developed for FastGPT v4.17.0. It covers configuration, constraints, and issue handling for tool calling and plugins in the declaration material and case file audit scenario of tendering and procurement compliance management.
What data for this category looks like
Data for this category covers end-to-end procurement documents across central state-owned enterprise and public sector procurement centers, tendering offices, business demand departments, audit and discipline inspection departments, and tendering agencies. Included documents are procurement plans, demand specifications, tender documents, bid evaluation records, contracts, acceptance documents, and supplier complaint materials. Data sources include enterprise procurement platforms, public resource trading platforms, supplier databases, contract management systems, and audit issue databases. Data is updated daily. Documents contain fixed fields such as procurement method, budget basis, scoring rules, supplier qualification requirements, and performance acceptance standards. Some documents include cross-department collaborative approval node identifiers.
What constraints these characteristics impose on the tool calling and plugins link
Data sources for this category are scattered, and update frequency is high. Tool calling must connect to both internal enterprise procurement systems and public resource trading platforms, and adapt to interface formats of different platforms. Document fields cover the full procurement compliance process. Tool calling must accurately match core compliance fields such as procurement method and budget basis, to avoid missing key verification items. Daily updated data volume fluctuates greatly. Tool calling must set reasonable concurrency thresholds to prevent interface overload. Error consequences of compliance auditing directly relate to audit and accountability. Tool calling output must include traceable data source identifiers, to ensure audit results can be traced.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
toolChoice | auto | Adapt to the automatic selection requirement of multiple types of compliance verification tools in tendering and procurement scenarios |
functionCall | Set based on actual testing | Support function call adaptation for Qwen2.5 models, must align with the native capabilities of the model |
| Maximum single tool call limit | First 3 entries | Avoid pulling too much data in a single call leading to timeout, match the number of core fields for single case file audit |
| Tool call timeout period | 600 seconds | Adapt to the delay characteristics of cross-platform data pulling, cover the interface response cycle of enterprise procurement platforms and public resource trading platforms |
| Plugin call retry count | 2 times | Respond to temporary fluctuations in public resource trading platform interfaces, avoid audit interruption caused by single call failure |
| Context window size | 8000 token | Accommodate full documents and historical approval records for a single case file, meet the context requirements of compliance auditing |
Three common configuration mistakes
- Symptom: Tool calling outputs both AI natural language replies and next-level branch execution logs, with both results displayed on the interface. Cause: The
toolChoiceparameter is not configured correctly, and multi-output mode is enabled by default. - Symptom: Chinese fields returned after MySQL plugin calls are garbled, and supplier names in procurement plans cannot be read normally. Cause: The correct character set parameter is not configured in the database connection plugin, and utf8mb4 encoding is not specified.
- Symptom: Calling the Qwen2.5 model to initiate a tendering method verification question does not trigger the tool calling logic, and only returns a natural language answer. Cause: The forced trigger mode of the
functionCallparameter is not enabled, and the model is not configured to give priority to tool calling.
How to confirm correct configuration
- Initiate a single case file audit test, trigger tool calling, check whether the output only contains tool calling results or natural language answers, with no simultaneous output.
- Call the MySQL plugin to import procurement plan documents, check whether Chinese fields are displayed normally, with no garbled characters.
- Switch to the Qwen2.5 model, initiate a question including tendering method verification, check whether tool calling is triggered and corresponding verification results are returned.
- Simulate an interface timeout scenario, check whether the plugin automatically retries the specified number of times, with no direct interruption of the audit process.
The values listed above are common starting points, and should be evaluated against local samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.