Treasure AI gives you four different ways to talk to an AI agent, and they are easy to confuse because three of them are built on the same underlying platform. Treasure AI Studio is a standalone, multi-platform chat application and the broadest of the four — it queries any data in your account, builds segments and full customer journeys, and generates files, all in one conversation. Foundry Workspace, Audience Agent, and Generic Chat are three narrower, purpose-built ways of running a single agent you configure in AI Agent Foundry. Picking the wrong one costs you time — for example, trying to schedule a conversational agent that only supports form-based runs, or looking for a "Foundry" tab inside Treasure AI Studio that doesn't exist. This page defines each surface, compares them side by side, spells out how Treasure AI Studio differs, and explains exactly how a custom Foundry agent connects to it.
| Surface | What It Is | Where You Access It | Primary Persona |
|---|---|---|---|
| Treasure AI Studio | A multi-platform chat application that queries any data in your account without SQL, builds CDP segments and full customer journeys, generates spreadsheets/documents/presentations/charts, and carries context across sessions with Work Folders. The broadest, general-purpose surface — not scoped to one parent segment or one pre-built agent, and a separate product from AI Agent Foundry. | agent.us01.treasure.ai and equivalent regional URLs, or the Desktop (Mac or Windows) / iOS apps | Marketer, Data Analyst, Project Owner |
| AI Agent Foundry | The platform where agents are built and governed: projects, knowledge bases, agents (system prompt + tools), integrations, and interfaces. Foundry Workspace, Audience Agent, and Generic Chat are all ways of running an agent defined here. | Treasure Console → AI Agent Foundry | Developer / Solutions Engineer building agents |
| Foundry Workspace | The runtime environment for form-type Foundry agents. Supports interactive runs and recurring schedules (daily, weekly, monthly). | Treasure Console → Foundry Workspace | Marketer or analyst running a recurring report or generation task |
| Audience Agent | A managed AI Agent Foundry agent, exposed through a Parent Segment integration, scoped to one parent segment's data. A premium feature — contact your Customer Success representative. | Audience Studio → a parent segment → Audience Agent | Marketer analyzing or segmenting one parent segment |
| Generic Chat | The AI Agent Foundry integration type (and its Treasure Console tab) that exposes any self-defined agent as a chat widget for testing or internal use. | Treasure Console → AI Agent Foundry → Generic Chat tab | Developer testing an agent; internal users of a custom agent |
A Foundry-based agent is a fixed pipeline: whoever built it wired its knowledge bases and tools to a system prompt, and every run follows that same wiring. Treasure AI Studio is a different kind of agent — it decides its own path for each request rather than following one someone else configured. Its Agent Orchestrator runs a five-stage loop on every message: assemble context (system prompt, your Work Folder's instructions, active Skills, conversation history), decompose the request into a plan, select which tool fits each step, execute and stream the results, then assemble a final response — repeating steps as needed until the task is done. Where a Foundry agent's tools are fixed at build time, Studio picks a different tool for step 2 based on what step 1 returned, and changes its plan mid-conversation if a query fails or you correct it.
| Capability | Treasure AI Studio | Foundry Workspace / Audience Agent / Generic Chat |
|---|---|---|
| How it decides what to do | Plans a multi-step approach per request and adapts based on each tool's result — see Agent Orchestrator | Follows the fixed prompt template and tool wiring the agent's builder configured; the same input takes the same path every time |
| Tools available in one session | tdx query/ps/sg/journey/wf, file read/write, web fetch and search, and connections to Databricks/Snowflake/BigQuery — the model chooses which to call and in what order | Only the specific Knowledge Base, Agent, Image Generator, or Workflow Executor tools the builder attached to that one agent |
| Domain competence | Skills teach the agent new competence — SQL patterns, segment schemas, workflow syntax — installed or built per Work Folder from the Skills Marketplace, and applied automatically by the orchestrator during planning and tool selection. No agent configuration to touch | Fixed at build time; adding a new capability means editing the agent's system prompt or tools in AI Agent Foundry |
| Scope of one conversation | Any database, segment, journey, or workflow in your account | Whatever one Foundry agent's knowledge bases and tools were configured for; Audience Agent is scoped to a single parent segment |
| Journey creation | Yes — analyze a segment, design a journey, and create it in the same conversation | Not available; Audience Agent stops at segment analysis and creation |
| File generation | Spreadsheets, Word documents, PowerPoint decks, charts, and code, each with an interactive viewer | Limited to what the agent's Output schema defines, plus a dedicated Image Generator tool for images |
| Context across sessions | Work Folders carry standing instructions (default database, conventions, prior decisions) into every new chat | Not documented — each agent's behavior is fixed by its system prompt at build time |
| Platforms | Web, Desktop (Mac and Windows), and Mobile (iOS) — one account, same capabilities everywhere | Browser only, inside Treasure Console |
This is why Treasure AI Studio is the right default for open-ended analysis, segment-to-journey work, and report generation — it plans its own way there instead of running a path someone else fixed in advance. Reach for a Foundry-based surface instead when you specifically need that fixed, governed path: Audience Agent to keep a marketer scoped to one parent segment, Foundry Workspace for a repeatable form a non-technical user runs on a schedule, or Generic Chat/Webhook/Slack to expose one fixed-purpose agent outside a full Studio session.
AI Agent Foundry is the one platform behind three of these surfaces. You build a project, define agents inside it, and then choose an integration or interface to decide where that agent shows up:
- A Parent Segment integration surfaces the agent as an Audience Agent inside a specific parent segment in Audience Studio.
- A Generic Chat integration surfaces the agent as a chat widget in the Generic Chat tab, inside Treasure Console.
- A Form-type interface surfaces the agent in Foundry Workspace, where it can also be scheduled.
- Webhook and Slack integrations surface the agent outside Treasure Console entirely.
See Creating an Integration for the full list of integration types and Creating an Interface for how form-type interfaces work.
Treasure AI Studio sits outside this model entirely — no project, agent, or integration to configure. See How Treasure AI Studio Differs above for what it runs instead.
Not as a live tool the Studio orchestrator calls during a chat. Creating an Integration documents exactly four integration types for a Foundry agent — Parent Segment, Generic Chat, Webhook, and Slack — and none of them targets Treasure AI Studio. Studio's own documented extension point is Skills, not AI Agent Foundry integrations.
Where the two products do connect is at build time, through Treasure Code: the tdx agent pull, tdx agent push, and tdx agent test commands let you author and validate an AI Agent Foundry agent — its system prompt, tools, and outputs — as version-controlled YAML and Markdown files. Treasure AI Studio ships the same workflow as three Skills you can activate in a work folder: foundry-agent (build with tdx agent pull/push), foundry-agent-prompt (write the system prompt), and foundry-agent-test (validate with tdx agent test). Using these Skills means you are using Treasure AI Studio as a development environment for a Foundry agent — the agent still runs in its Foundry surface (Foundry Workspace, Audience Studio, Generic Chat, Webhook, or Slack), not inside the Studio chat session that built it.
An agent built in AI Agent Foundry can call another agent as a tool — this is how you compose a setup like a parent segmentation agent that delegates a specialized task to a custom agent you built. Creating an Agent documents this as the Agent target for a tool: you select a target agent, choose an output mode (Return — the caller uses the target's output to build its own response — or Show — the target's output is rendered directly to the user), and optionally pass arguments through a JSON schema. In a tdx agent pull/push YAML file, the same thing is a type: agent tool entry referencing the target with @ref(type: "agent", name: "...") — see Agent Commands — Tools Configuration.
This orchestration runs inside AI Agent Foundry and whichever surface exposes the top-level agent (Audience Agent, Generic Chat, Foundry Workspace, Webhook, or Slack) — it is not something you configure from, or trigger through, Treasure AI Studio.
| Capability | Where It Lives |
|---|---|
| Run an agent on a recurring schedule | Only Foundry Workspace, and only for form-type agents (daily, weekly, or monthly). See Scheduling Agents in Foundry Workspace. Audience Agent and Generic Chat conversations cannot be scheduled. |
| Trigger a Treasure Workflow from an agent | Any AI Agent Foundry agent, via the Workflow Executor tool (target function Execute workflow, or Search workflow to discover one first). See Creating an Agent — Workflow Executor. |
| Permission required to use Generic Chat or Foundry Workspace | The Generic Chat in LLM Console permission, part of the General LLM Access policy, for each relevant LLM project. Without it, the Generic Chat tab and Foundry Workspace are not available. |
Is Audience Agent being replaced by Treasure AI Studio? No product documentation marks Audience Agent as deprecated. Both are currently documented, active surfaces: Audience Agent is scoped to analyzing and segmenting a single parent segment inside Audience Studio; Treasure AI Studio additionally builds journeys, generates files, and handles general-purpose tasks in the same conversation. Choose based on the task, not on an assumption that one replaces the other.
Can I build a multi-agent setup with a segmentation agent and a custom agent I wrote? Yes, inside AI Agent Foundry — add an Agent tool to the top-level agent, targeting your custom agent, then expose the top-level agent through the integration you need (Parent Segment for Audience Studio, Generic Chat for Treasure Console). This is a Foundry-native capability; it does not route through Treasure AI Studio.
Why can't I type anything in Generic Chat? This is almost always a missing permission. Confirm you have Generic Chat in LLM Console in the General LLM Access policy for that project.
Can an agent create a segment automatically? An agent with a Parent Segment Knowledge Base tool can query and analyze parent segment data conversationally (see Using Audience Agent to Create a Segment), but actual segment creation and definition still go through Audience Studio's segment tools, not a fully autonomous agent action.
- AI Foundry Overview — the project/agent/integration model in full
- Foundry Workspace Overview — running and scheduling form-type agents
- Audience Agent Overview — segmenting one parent segment conversationally
- Treasure AI Studio — the multi-platform chat application and its own Agent Orchestrator
- Agent Commands — the
tdx agentbuild/test workflow that bridges Treasure Code, Treasure AI Studio Skills, and AI Agent Foundry