Built an in-sidebar knowledge base chat letting agents query past ticket intelligence directly from Zendesk

A conversational interface — built as a ZAF app and embedded in the Zendesk ticket sidebar — that lets support agents query the past-ticket knowledge base in real time without leaving the ticket. The chat is grounded in the same RAG retrieval pipeline that powers the First Response Agent, so agents and the AI share a single source of truth.

Where the First Response Agent produces a one-shot draft and the Feedback App captures evaluation, the Chat App is the exploration surface: agents ask follow-up questions, drill into historical resolutions, or check whether a similar ticket exists before responding.

Every answer streams back token-by-token over SSE and carries citation deep-links back to the source tickets it was grounded in, so an agent can verify a claim in one click rather than trusting the model blindly. It's built on the same Zendesk Apps Framework v2 foundation as the Feedback App, so both ship and update as part of the same suite.

I built the chat sidebar from the ground up and connected it to the existing past-ticket knowledge base, so the chat and the AI's first-response draft share a single source of truth.

I designed the citation experience that links every answer back to the original source ticket, plus the per-ticket session lifecycle so context stays scoped to what the agent is currently looking at.

I also implemented the SSE streaming layer so responses render incrementally in the sidebar instead of appearing all at once, keeping the chat responsive even on longer retrieval-augmented answers.