[Feature Request] Optional verified-answer mode for local Qwen models

I use Qwen models locally, and I think Qwen-Agent could provide something that is still missing from most local AI systems: an optional mode designed specifically to improve the reliability of normal answers.

Qwen-Agent already provides the foundation for tool use and web search. What I am suggesting is a stricter answer pipeline built around three steps:

Decide automatically whether a question requires web verification.

The user should not have to enable a research mode manually every time a question may depend on current or external information.

Evaluate the quality of the evidence.

The system should prefer primary and authoritative sources, consider publication dates, distinguish independent confirmation from repeated reporting, and retrieve the full page when a search snippet is not enough.

Require final verification before returning the answer.

The important factual claims in the draft should be checked against the collected evidence. When a claim is unsupported, contradicted or based on insufficient evidence, the system should be required to search again, correct or remove the claim, qualify the uncertainty, or abstain.

The key point is that verification should be a real gate, not another instruction that the model may ignore.

This would not require changing or retraining the Qwen model. The purpose would be to make the complete Qwen system produce better factual answers by combining the model with retrieval, source evaluation and mandatory checking.

I would especially like this to work with local Qwen models through OpenAI-compatible endpoints, using the same model for answering and verification when the user only has enough VRAM to run one large model.

Would an official verified-answer mode fit the direction of Qwen-Agent?