📄 中文摘要
在 n8n 工作流中，Claude 负责意图分类，Mistral 草拟响应，而 GPT-4 则进行最终润色。然而，当客户询问某个特定时间点的 AI 输出时，如何证明该输出未被篡改成为一个难题。多模型链条的每一步都是一个黑箱，虽然可以追踪输入和输出，但缺乏加密证明来确保它们之间的关联性、未被篡改及时间戳。这一问题是每个异构 AI 管道最终都会面临的挑战，主要体现在可验证性方面，而非性能或成本。
将 Claude、Mistral 和 GPT 集成到一个 n8n 工作流中 — 如何认证完整链条
📄 English Summary
Claude + Mistral + GPT in one n8n workflow — how to certify the full chain
In an n8n workflow, Claude classifies intent, Mistral drafts responses, and GPT-4 provides final polishing. However, when a client asks for proof of what the AI said at a specific step on a certain date, the challenge of verification arises. Each step in a multi-model chain acts as a black box; while the input and output can be tracked, there is no cryptographic proof linking them, ensuring they are untampered and timestamped. This issue of verifiability, rather than performance or cost, is a common challenge faced by heterogeneous AI pipelines.