Summary
K3 makes every customer↔rep↔agent conversation searchable and reviewable through Graph RAG, with citations — a second, rep-scoped corpus with index-time privacy and retraction paths.
History is the product
In most sales software, the conversation is exhaust — logged somewhere, read never. OpenKuai treats it the other way round: every customer↔agent/human conversation is logged, searchable and reviewable, by humans and by prime-agents, through Graph RAG with citations.
What it takes
Conversation retrieval is a different corpus from product knowledge, and the design says so:
- A second database — the conversation corpus is its own store, not an extension of the product KB.
- Rep-scoped access — a rep’s searches reach their conversations and their team’s; a manager’s reach their team’s.
- Privacy at index time — a conversation involves a third party who never consented individually, so the scope of what enters the index is data design, and
never-touchneeds both an index-time path and a retraction path.
Why agents need it too
The same index serves the agent tier. A prime-agent answering a customer can search how similar enquiries were handled — with citations — instead of improvising. The conversation corpus is what makes the fleet of rep-paired agents collectively smarter than any single one.