Show HN: Product analytics (and evals) for agent sessions on your MCP

Armature这家公司推出了一款专为AI智能体打造的“用户行为分析神器”，堪称AI会话界的“谷歌分析”。以前企业给产品加上AI助手后，用户和AI之间的对话仿佛黑盒，发生了什么全靠猜；而Armature能完整记录AI与用户交互的每一步，包括用户的真实意图、AI的思考过程，以及最终任务是否成功完成。通俗来说，就像给AI配了个“高清摄像头加记事本”。它能自动把用户的高频需求分类归档，比如“创建发票”、“对账”各占多少比例，还能揪出AI“卡壳”的环节——哪怕系统显示一切正常，它也能发现AI在暗中“原地打转”。更难得的是，连用户想做但AI没支持的“隐藏需求”也能被发掘出来。对企业而言，这项技术无疑是把“AI用户需求图谱”直接摆在眼前，让产品迭代从“拍脑袋”变成“看数据”，少走弯路；对普通用户来说，虽然看不见这个工具，但当AI越来越懂你、办事不再绕圈子时，服务体验的提升便是最直接的收获。

Product analytics for agent sessions See how users experience your product through Claude. ChatGPT. Claude Code. User sessions on your Claude Connector ChatGPT App or MCP live inside their AI client, not your UI. Armature captures and shows you what users ask, what agents think and how they use your product. Start for free → Book a demo Backed by Before Armature With Armature list_customers◂ create_subscription◂ send_invoice402 ◂ send_invoice◂ search_customers◂ get_plans◂ list_invoices◂ update_plan◂ usage_report◂ User intent · 14:16 Set up a paid plan for ACME and invoice them ✦ Agent thinking · 14:16 Invoice needs a billing contact — retrying with the owner email. ✓ Agent completed user task successfully 92score Before Armature Scattered tool call logs With Armature Session rebuilt User intent captured Agent thinking captured Session evaluated Use cases Learn your users' top use cases Armature's models read every session and identify what the user came to do. Sessions are grouped into use cases and ranked by volume and success rate, including the use cases you don't support yet. Create & send invoices38% Reconcile payments22% Bulk refunds not supported yet14% Export revenue report9% Issues Spot the top issues and fix them Models scan every session for failures, loops and dead ends, group them by root cause and rank them by how many users encountered them, even when every API response was 200 OK. 127 Agent loops on missing auth scope↑ 43 this week fix → 64 Search misses "refund" phrasing 31 Export truncated by pagination 12 Rate limit hit on bulk updates Session replay Replay and investigate any session Every session is scored on whether the user got what they asked for. Open any of them and replay the full trace: the ask, the agent's thinking and every call. When something breaks, you see exactly where. User intentSet up a paid plan for ACME and invoice them list_customers◂ ✦ Agent thinkingInvoice needs a billing contact — retrying with the owner email. send_invoice402 ◂ send_invoice◂ ✓ Agent completed user task successfully92score