Built on an MCP-native core so the AI, the UI, and your integrations share one honest surface.
Not a chatbot bolted on — an orchestrator that routes, drafts, and resolves with its own specialists.
Every message is fronted by an orchestrator agent. It understands intent, picks the right specialist, and keeps the user-facing voice consistent.
Triage, draft, knowledge, and admin specialists each have their own prompt, toolset, and model budget. Smaller jobs run on cheaper models — margins stay sane.
Spam, sentiment, intent, urgency, and language — one structured-output call per ticket. Intent uses your taxonomy, edited via chat. No per-seat AI fee.
When the AI is confident from your KB, it sends the reply and closes the ticket. Opt-in, configurable threshold, citations on every auto-reply.
When solved tickets cluster on an intent with no published article, we surface a draft built from the actual resolutions. Review and publish, or dismiss.
Hybrid search fuses full-text and vector retrieval, so drafts cite the right article instead of inventing one.
Tone, forbidden phrases, and brand guidance flow into every reply. Keep it warm, concise, formal, or playful — per workspace.
Every ticket carries an AI-spend ceiling. A runaway loop, a malicious chat visitor, or a bad prompt can't burn meaningful margin on your dime.
Side-conversations let the agent loop in another team or vendor on a ticket without ever involving the customer. Internal threads, never sent.
All the workspace primitives, none of the ceremony.
One unified queue. Status, priority, assignee, group, SLA — all where you'd expect them.
Filter by anything the rules DSL understands. Built-ins cover the basics; ask the agent for a custom view.
First-response and resolution targets with per-calendar business hours, pauses, and breach actions.
Event-driven rules — classify, assign, auto-reply — running inline with a full cause trail.
Time-based reactions: escalate after 4 hours, close stale pendings, chase missing info. All through the same DSL.
Templates for the things you say a hundred times. Drop {{requester.first_name}} placeholders and they fill in. Render-and-review, never auto-send.
Email and chat, treated as first-class — same data model, same agent, same audit log.
Inbound parsing, outbound through your domain, threading that actually holds. CC, BCC, and signatures.
JS snippet for your site. Routes through the same KB-grounded AI as email; hands off to a human when it can't answer. Brand, color, position configured via chat.
Everything you can click, you can script. External MCP servers plug straight in for custom integrations.
Explorer is built for the questions teams actually ask — not for a dashboard beauty contest.
Resolutions, SLA performance, volume, knowledge health — with drill-downs that match how we bill.
Ask the agent in plain language; it builds a saved report you can reopen any time.
Every mutating action — human, AI, or system — is logged with actor, subject, and cause. Filterable, exportable.
The boring, load-bearing stuff done right.
Same tool surface for the UI and the AI. No "internal-only" shortcuts that bypass the rules.
Install first-party packs or point the workspace at any MCP server. Per-app capability allow-lists.
Postgres row-level security enforced at the DB, not the app. No cross-tenant code paths.