# Chat Overview Chat is the working surface of Treasure AI Studio: you describe an outcome in plain language, and an AI agent runs the work — querying your Treasure Data CDP, building segments, drafting journeys, generating files — with every action visible in the stream as it happens. It replaces the "learn the tool first, then get the answer" order of a console with a conversation, so a marketer can reach a cart-abandonment audience in minutes and an analyst can go from question to validated SQL without leaving the thread. In platform terms, chat is where a human sets the direction and the guardrails while AI super agents run the Customer Intelligence Loop — the cycle of collecting, unifying, understanding, deciding, and engaging that Treasure AI closes in minutes instead of weeks. Chat is not a chatbot layered over a dashboard. Each conversation gets its own server-side session with real tools, real credentials handled at the platform boundary, and a transcript you can read back — which is why the pages below spend as much time on sessions, memory, and history as on typing a message. AI Chat Interface Send messages, read streaming responses and tool calls, invoke skills with `/`, and attach files. Chat Sessions & Limits Status labels, the time limits that end a session, and what the AI remembers when a chat resumes. Chat History How conversations are saved server-side, titled automatically, searched, renamed, and deleted. In-Session Memory How the context window fills, what automatic compaction preserves, and how to structure long work. Voice Input Dictate messages instead of typing — and what differs between Web, macOS Desktop, and iOS. ## Frequently Asked Questions ### What Can I Ask Studio to Do? Treasure AI Studio handles the work you would otherwise split across a SQL console, the CDP UI, and a spreadsheet: exploring databases and schemas, running Trino queries, building parent and child segments, designing journeys, generating charts and documents, and reaching external sources you have connected. Requests that need current outside information — market trends, competitor news — pull in a web search on their own. Studio's [skills](/products/ai-studio/skills/skills) are what teach it your platform's specific patterns, so the catalog of skills is a good map of its range. ### Does the AI Touch My Real Customer Data? Yes — that is the point, and it is why the boundaries are explicit. The agent runs in an isolated server-side sandbox and queries your account through the same permissions you have, without ever holding your real credentials: the platform substitutes those at the network boundary. Every outbound request is recorded in the [Network Audit Log](/products/ai-studio/security/network-audit-log), and administrators can restrict where the sandbox may connect with a [network policy](/products/ai-studio/security/network-policy). See [Security Overview](/products/ai-studio/security). ### How Long Does a Conversation Stay Available? **28 days from the day it was created, by default.** Every Treasure AI Studio chat expires: an administrator sets the retention period to 14, 28, or 60 days, and no setting keeps chats indefinitely. Until then, messages are saved server-side as they are sent, so a chat is available from any device you sign in on — and you can delete one sooner yourself, or lose it by deleting the work folder it belongs to. Because the clock runs from creation rather than your last message, anything you need to keep belongs in a [work folder](/products/ai-studio/concepts/work-folders) file, which retention does not delete. See [Chat History](/products/ai-studio/chat/chat-history) and [Chat Data Retention](/products/ai-studio/security/data-retention). The *working session* behind a chat is a different thing and much shorter-lived: it is released after inactivity and restarts on your next message ([Chat Sessions & Limits](/products/ai-studio/chat/sessions)). ### Do Long Conversations Forget Things? Studio compacts a conversation automatically when it approaches the model's context limit, summarizing older exchanges and keeping the decisions, results, and definitions that matter. You never run a command to make that happen, and a **"Context compacted"** marker plus an expandable summary shows exactly what was carried forward. To carry context across separate chats rather than inside one, use a [work folder](/products/ai-studio/concepts/work-folders) — its instructions file is read at the start of every chat in that folder. See [In-Session Memory](/products/ai-studio/chat/in-session-memory). ### What Does a Chat Cost? Chat consumes AI credits, and the amount depends on the model tier the conversation is using: **Fast** for quick lookups, **Balanced** by default, **Pro** for multi-step analysis. You can switch tiers mid-conversation from the composer, and administrators can cap daily credit use per user — when a cap is set, the composer shows your usage for the day. See [AI Model Tiers](/products/ai-studio/concepts#ai-model-tiers), [Credit Usage](/products/ai-studio/credits/credit-usage), and [Credit Policy](/products/ai-studio/credits/credit-policy). ## Next Steps - [Getting Started](/products/ai-studio/getting-started) — Install and sign in on Web, Desktop, or Mobile - [Core Concepts](/products/ai-studio/concepts) — Work folders, model tiers, credits, and the agent orchestrator - [Agent Orchestrator](/products/ai-studio/concepts/orchestrator) — How a request becomes a plan, tool calls, and a result