Helpway v1.2 — Smarter RAG, lower latency, new MCP server

Hello! Quick update on what shipped this month.

Smarter RAG. We rebuilt the retrieval layer on top of Cloudflare Vectorize with BM25 hybrid scoring. Recall on customer KBs improved ~22% in A/B.

Lower latency. First-token from Tokyo is now 480ms (was 620ms). Preparing streaming payloads on the edge is faster when we skip the ORM for known hot paths.

New MCP server. Point Claude Code at your Helpway workspace and it can read every conversation, every article, and every tool call.