> ## Documentation Index
> Fetch the complete documentation index at: https://docs.upsolve.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Workspaces
> One data model and one agent, versioned and published together — the surface where you build and ship an analytics agent.
## What is a Workspace?
A Workspace is where you build one analytics agent. It combines, in a single versioned surface:
* **The Data Model** — which tables and columns the agent can see, what they mean, and who can query what (row-level security)
* **The agent's context** — knowledge base files, skills, system prompts, per-user memory, and golden assets
* **Testing** — a Playground where you chat with the agent as one of your project users
* **Deployment** — an embed snippet and preview links for putting the agent in your product
The Data Model and the agent are no longer separate concepts with separate version histories. A workspace *is* its data model — you create one workspace per agent, and the workspace publishes the schema and the agent's context together, so the two can never drift apart.
## Creating a Workspace
Workspaces live inside a project. Open your project and you'll see **Your Workspaces** — the list replaces the separate Data Models and Agents tabs.
1. Click **New workspace**
2. Enter a **name** (and an optional description)
3. Select the **connection** to base it on
4. Click **Create workspace**
A connection is required and cannot be added later. The workspace's schema is read from it, so a workspace without one has nothing to model. If your project has no connections yet, add one in Project settings first.
Creating the workspace introspects the connection immediately, so its catalog of tables and columns is ready to curate right away.
## The Workspace Sidebar
Opening a workspace swaps the app sidebar to the workspace's own sections, grouped by the part of the lifecycle they belong to:
| Section | What it's for |
| ------------------ | ------------------------------------------------ |
| **Data Model** | Tables, columns and row-level security |
| **Knowledge Base** | Context files the agent can search and read |
| **Skills** | Reusable, on-demand know-how |
| **System Prompts** | Instructions the agent runs under |
| **Memory** | What the agent remembers about each user |
| **Assets** | Golden questions the agent must answer correctly |
| **Playground** | Chat with this workspace's agent |
| **Deploy** | Embed this workspace in your product |
| **Chat History** | Every conversation in this workspace |
The first six are the context you give the agent; Playground and Deploy are how you try it and ship it; Chat History is the record of what it did. Workspace settings sits at the foot of the sidebar rather than among the pages.
The header is a breadcrumb — **Project / workspace** — and both steps are switchers: the workspace picker also offers **Create workspace**, and the project picker **Create project**, so you never have to navigate back to make the next thing.
## Draft, Publish and Production
The version bar lives on the workspace, above every page, and it is the one place that says where your work stands:
* **Draft** — every page edits the same working copy, and edits **autosave**. There is no Save button; the version selector shows when the draft was last written.
* **Publish** — stages the draft as a new numbered version. Publishing captures the schema, the prompts, the skills and the assets together.
* **Promote to production** — a published version can be promoted, and production is what deployed users and embeds are served. This replaces the old "set this data model version / agent version as production" two-step: one workspace, one production switch.
* **Discard / Reset** — throw away draft changes and return the working copy to the published version.
Selecting a published version in the bar makes every page read-only for that version, so you can inspect exactly what shipped. **View changes** beside the selector diffs the draft against the latest published version across the whole workspace.
## Deploying a Workspace
The **Deploy** page produces everything you need to put the agent in your product:
1. **Set up your frontend** — a one-time setup per product, with the backend and frontend guides.
2. **Embed the iframe** — point it at `/share/workspace/{projectId}/{workspaceId}?jwt=`, passing a short-lived project-user token minted by your backend.
3. **Preview it as one of your users** — generate a real, short-lived preview link (valid for one hour) and open the embed exactly as a project user would see it, before writing any integration code.
A status strip at the top answers the first question you'll have: which version the embed currently serves. Until a version is production, the embed has nothing behind it.
See [Deploy & Expose](/ai-agent-builder/deploy-agents/overview) for the full setup.
## Legacy Agents
Projects that were built before workspaces keep their existing agents on a legacy page, where they remain fully readable and editable. New agents cannot be created there — new work should happen in a workspace, where the data model and the agent ship together.
## Next Steps
* [Curate the workspace's Data Model](/ai-agent-builder/data-models) — select tables, add descriptions, and configure security
* [Encode Your Context](/ai-agent-builder/encode-your-context) — prompts, golden assets, skills, knowledge base and memory
* [Follow the Complete Setup Guide](/ai-agent-builder/setup-guide) — from a blank project to production