Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Volcano Engine APMPlus adds Agent trajectory observability

AI summary

Volcano Engine APMPlus now provides an Agent trajectory view that connects user messages, Agent responses, tool calls, and returned results in interaction order. Developers can inspect step-level inputs and outputs, latency, Token consumption, tool status, and Subagent branches to diagnose incorrect answers, slow execution, or abnormal Token usage. The capability complements Trace, which organizes the same execution as a Span-based call tree.

In-depth insight

APMPlus Agent trajectories give developers a source-supported way to investigate cases where tool calls succeed but the final answer is wrong, linking the discrepancy to specific prompts, context, tool results, or response steps.

01导语

对 Agent 开发者和使用者来说,最棘手的往往是:任务显示“已完成”,但答案不对、关键信息遗漏;或者同样的任务突然变慢,Token 消耗也涨了。只看最终输出和总耗时,很难判断问题发生在哪一步,更难决定该调整提示词、工具调用,还是上下文处理。

在《APMPlus・看得见的 Agent》前两篇文章中,我们分别介绍了 OpenClaw 和 Hermes 的观测数据接入。模型、工具等调用记录进入 APMPlus 后,开发者可以沿着 Trace 查看调用关系、输入输出、耗时和 Token 消耗,排查报错与性能问题。

但调用成功,不代表回答就符合预期。比如,用户问“老用户能参加这次活动吗?”,Agent 成功调用了查询工具,返回规则里写着“仅限新用户”,最终却回答“可以参加”。整个过程没有报错,要查清原因,就得核对:工具拿到的规则是否完整,后续模型调用是否收到这条限制,最终回答又从哪里偏离了规则。

#Volcano Engine#APMPlus#APMPlus Agent轨迹#OpenClaw#Hermes

Original · 10 min readRead the original
Group chat

No comments yet. Start the conversation.