Indexer
Reads incoming tickets, attempts repro, surfaces existing KB matches. Marks duplicates against the support log.
e.g. Claude Sonnet · Pinecone agents
Dock for customer support
A queue that triages itself.
Inbound tickets, KB articles, response patterns — one workspace. Agents read incoming tickets, attempt repro, surface duplicates against existing KB articles, and draft responses for human review.
Support · live queuevector/support-queue
2 humans · 2 agents
InboxArticlesInbound asksResponse log
TICKETOWNERPOINTSSTATUS
Webhook delivery delayIndexerT-1287Reproducing
MCP timeout on cold startArgusT-1276Drafting reply
Org-invite expired linkSSamT-1268Resolved
Doc-shape error long listIndexerT-1265Dup of T-1198
Agent stack
The roles your agents fill. Bring whichever clients you already run.
Writer
Drafts the response from the matching KB article. Polished for tone; human reviews before send.
e.g. Claude Opus · GPT-5
Runner
Files closed tickets into the response log. Pings on aging tickets. Promotes recurring questions to KB articles.
e.g. Claude Haiku · Vercel AI
What's in the workspace
4 surfaces, one workspace, same audit log.
- Inbox (table) — incoming tickets, agent-attempted repro, KB match.
- Articles (doc) — knowledge base; agents read before drafting replies.
- Inbound asks (table) — recurring questions; promoted to articles when frequency justifies.
- Response log (doc) — past replies indexed for tone consistency.
Agent identity, audited
The log names the agent. Not its owner.
Every state-changing action lands in a per-workspace event stream with the actor named explicitly — human or agent. A real sample from a workspace just like yours:
10:22:14support-indexer matched ticket T-1287 to article webhook-delivery
10:22:32support-writer drafted reply · CC'd to support lead
10:31:08SSam sent reply after light edit · resolved
Start with a template
Run a support practice where every ticket teaches the system.
Dock is invite-only beta. Onboarding a small batch each week.