Added
- @ReplyMill mentions in Slack — type
@ReplyMill <question>in any channel where the bot is a member and get an AI-synthesized answer from your team's actual support data, posted as a threaded reply with clickable sources.
What you can ask
- "What is the most popular feature request?" — ranks your aggregated request clusters by demand score (unique customers × volume × recency) and surfaces the top one with a representative thread.
- "Are there any common bugs?" — groups bug-labeled threads by topic and reports the most frequent, with a representative thread per topic.
- "How did we handle refund disputes?" — semantic search over your past customer threads, paraphrase-tolerant.
- "How many open p1 issues do we have?" — aggregate counts by category, priority, and resolved state.
- "Show me unresolved feature requests from last week" — structured filters on category, priority, resolved status, and time window.
Compound questions like "What's our top request and how have we been responding to it?" trigger multiple lookups in a single reply.
Improved
- Every reply is scoped to your company's data only — cross-workspace queries are structurally impossible.
- Sources cited inline as
[#1],[#2]link directly to the corresponding thread in your ReplyMill inbox. - Answers stay attached to the conversation — replies post inside the thread you mentioned the bot in, so follow-ups stay in context.
Under the hood
- Semantic search runs over your full thread history with pgvector and incremental embedding refresh on every new message — new conversations are searchable within seconds.
- Five-tool agent loop (semantic search, request clusters, bug topics, structured filters, aggregate stats) running on
gpt-4o-mini. Most answers return in 2-4 seconds.