Creating Workspaces

Workspaces are the core of Roundtable. Each workspace is an isolated AI environment where your team collaborates with an AI model — complete with its own compute resources, storage, database, and tools.

What Is a Workspace?

A workspace is a fully isolated environment that runs in its own Kubernetes pod. This means:

• Dedicated compute — Each workspace gets its own CPU and memory allocation, not shared with other workspaces.
• Isolated storage — Files, code, and data in one workspace are completely invisible to other workspaces.
• Own database — Conversation history, tool state, and workspace metadata are stored independently.
• Full tool access — All 16 built-in tools (code execution, web search, file management, database queries, etc.) run within the workspace's isolated environment.
• Independent AI configuration — Each workspace can use a different AI provider, model, and system prompt.

Think of a workspace as a private room for your team and an AI — everything that happens inside stays inside.

How to Create a Workspace

1. From the Dashboard, click New Workspace.
2. Fill in the configuration options (see below).
3. Click Create.

Roundtable provisions a new pod on the cluster, which typically takes a few seconds. Once the workspace reaches the Running state, you can start chatting.

:::info Workspace limits The number of workspaces you can create depends on your plan. All plans support multiple workspaces — each workspace is billed at the per-workspace rate for your tier. :::

Configuration Options

When creating a workspace, you'll configure the following:

Name

A human-readable name for the workspace (e.g., "Frontend Team", "Data Pipeline Debug", "Q3 Planning"). This is displayed in the dashboard and in workspace headers.

AI Provider

Choose which AI provider powers this workspace. Roundtable supports five providers:

Provider	Description
Vertex AI	Google's enterprise AI platform. Uses service account credentials configured at the org level.
Google AI Studio	Google's developer-focused Gemini API. Requires an API key.
OpenAI	Access to GPT models. Requires an API key.
Anthropic	Access to Claude models. Requires an API key.
Ollama	Self-hosted open-source models. Connects to your own Ollama instance.

See AI Providers for setup instructions for each provider.

Model

Select the specific model to use within your chosen provider. Available models depend on the provider:

• Vertex AI — gemini-2.5-pro, gemini-2.5-flash, and other Gemini models
• Google AI Studio — gemini-2.5-pro, gemini-2.5-flash
• OpenAI — gpt-4o, gpt-4o-mini, o3, and others
• Anthropic — claude-sonnet-4, claude-haiku-3.5, and others
• Ollama — Any model available on your Ollama instance

You can change the model at any time after creation.

System Prompt

An optional set of instructions that shape how the AI behaves in this workspace. The system prompt is sent at the beginning of every conversation and sets the context for all interactions.

Examples:

You are a senior backend engineer. Help the team debug Node.js services,
review API designs, and write integration tests. Always explain your reasoning.

You are a data analyst. When asked questions, prefer querying BigQuery for
real data over making assumptions. Format results as tables when possible.

tip

A well-crafted system prompt can dramatically improve the quality of AI responses for your team's specific use case. Be specific about the AI's role, preferred tools, and output format.

Workspace Lifecycle

Every workspace moves through a set of states during its lifetime:

┌──────────┐     ┌──────────┐     ┌──────────┐
│ Running  │────▶│ Standby  │────▶│ Stopped  │
└──────────┘     └──────────┘     └──────────┘
     ▲                │                │
     │                │                │
     └────────────────┘                │
     └─────────────────────────────────┘

State	Description
Running	The workspace pod is active and ready. Team members can chat, use tools, and collaborate.
Standby	The workspace has been idle and is preparing to scale down. It will wake automatically on access.
Stopped	The workspace pod has been fully deallocated. No compute resources are consumed. Can be restarted manually or automatically.

Transitions

• Running → Standby — Triggered automatically after 30 minutes of inactivity (no messages, tool usage, or API calls).
• Standby → Stopped — The workspace completes its scale-down process and the pod is deallocated.
• Stopped → Running — Triggered when any user accesses the workspace (opens it in the dashboard or sends a message). The pod is reprovisioned automatically.
• Running → Stopped — A workspace can also be stopped manually by an admin or owner.

Scale-to-Zero

Roundtable uses a scale-to-zero model to keep costs predictable and avoid wasting resources on idle workspaces.

• After 30 minutes with no activity, the workspace enters Standby and then Stopped.
• When a team member opens the workspace or sends a message, the pod is automatically reprovisioned. This takes a few seconds.
• Conversation history, files, and settings are preserved across sleep/wake cycles — nothing is lost.
• Stopped workspaces consume no compute resources and do not count toward CPU or memory limits.

info

Scale-to-zero is fully automatic. You don't need to configure anything — workspaces sleep and wake on their own. You can also manually stop or start workspaces from the dashboard if needed.