> ## Documentation Index > Fetch the complete documentation index at: https://docs.thread.app/llms.txt > Use this file to discover all available pages before exploring further. # Concierge chat > The Account Concierge is a streaming AI chat on every portal page that answers from the account's own documents, schedules time with your team, and forwards customer uploads to you. Every page of the customer portal carries a floating chat bubble (with your logo inside it) that opens the **Account Concierge** — an AI assistant scoped to that one account's onboarding. It answers from the documents and knowledge on the account, helps the customer book time with your team, and takes in files the customer wants to hand over. The concierge's AI nature is disclosed structurally: the chat header always shows "AI assistant for" your brand name, on every render. The agent's name — Account Concierge — is fixed; your brand is what wraps it (the logo in the bubble and header, the brand accent, the sanitized prose). ## What it can answer The concierge is deliberately bounded to onboarding — it's not a support, sales, or billing agent, never quotes pricing, and never makes commercial commitments. Within that scope it can: * **Answer from the account's documents** — it searches the documents shared with the account, the customer's uploads, and the account's knowledge, and can pull up the SOW and its status. * **Explain where things stand** — what's left to set up, what's needed from the customer next. The empty state offers starters like "What do you need from me next?" and "Walk me through what's left to set up." * **Schedule time with your team** — it proposes real open slots from the assigned teammate's calendar and the customer picks one with a tap. It can never free-book a time the customer states; propose-then-pick is the only path. * **Bring your team in** — it can notify the assigned teammate, escalate to the team, request a callback, or send over a summary of the conversation. These land with you in Slack or email. ## How a conversation reads * **Streaming** — replies stream in token by token; while the concierge is working, a quiet shimmer shows instead of a spinner. * **Tool work shows as one warm line per step**, never raw output — "Searching your account docs" resolving to "Found 4 relevant passages," "Checking your SOW," "Letting Sam know." A failed step reads "That didn't go through." * **Every message is timestamped** — a relative time under each bubble ("2 minutes ago"), with the exact date and time on hover. ## Uploads forwarded to your team Customers can attach files to the chat (or drag them onto it). The concierge reads the file and drafts a hand-off: a summary plus any extracted fields, presented back to the customer as an editable review card. The file is stored privately against the customer's portal identity; the concierge reads it ("Reading what you uploaded"). A review card shows the drafted summary and extracted fields. The customer can edit the summary before sending — nothing is forwarded without their confirmation. On "Send," the file is bridged into the account's documents on your side and the assigned teammate is notified in Slack (falling back to email). The notification is footed with the AI disclosure — composed by the Account Concierge, your brand's AI assistant. ## Where you see it Conversations, escalations, and forwarded uploads surface to your team through your existing channels — Slack notifications, email, and the account's documents. Scheduling requests resolve against the assigned teammate's connected calendar, so a booked slot is a real event. Where forwarded uploads and shared documents live on both sides. For structured, multi-field asks, the concierge pattern scales up to a full guided flow.