Creating a New Workspace

When I start a new AI evaluation project, I want to create a dedicated workspace, so I can keep my research organized and separate from other initiatives.

Introduction

Context pollution is a constant battle for research engineers juggling multiple infrastructure decisions. Your Claude vs GPT-4 evaluation gets mixed with your GPU pricing research. Your fine-tuning notes bleed into your RAG architecture investigation. Six months later, you’re searching through a tangled mess of sources and artifacts.

Lattice workspaces provide complete isolation for each research context. Think of a workspace as a dedicated research environment—its own sources, its own chat history, its own artifacts. When you’re deep in GPU memory calculations for Llama fine-tuning, your inference latency research stays completely separate.

Step 1: Open the Workspace Selector

The Lattice interface features a three-panel layout: Sources on the left, the Lab (chat) in the center, and Studio (artifacts) on the right. At the top of the Sources panel, you’ll see the workspace selector showing your current workspace.

Click the workspace selector to manage your workspaces.

Step 2: View Existing Workspaces

The dropdown reveals all your workspaces. Each entry shows:

• Workspace name (e.g., “Nvidia Research,” “LLM Architectures”)
• Source count — how many documents are indexed
• Message count — your conversation history

A checkmark indicates the currently active workspace. At the bottom, you’ll find the ”+ New Workspace” button.

Step 3: Start Creating a New Workspace

Click ”+ New Workspace” and an inline input field appears. This keeps you in flow—no modal dialogs or page navigations. Simply type your workspace name directly.

Step 4: Name Your Workspace

Enter a descriptive name that reflects your project’s focus. Use naming conventions that help you quickly identify workspaces later:

• “H100 Training Stack - Q4 2024”
• “Inference Provider Comparison”
• “RAG Architecture v2”
• “Claude Sonnet Evaluation”

Step 5: Your New Workspace is Ready

Press Enter to create the workspace. Lattice immediately:

• Creates the workspace and switches to it
• Updates the selector to show your new workspace as active
• Displays the empty workspace state with “No sources added yet”
• Generates fresh Suggested prompts relevant to a new project

Workspace Isolation

Each workspace is a complete isolation boundary:

workspace_1 (H100 Training Stack)
├── sources[]        # Training papers, GPU specs
├── messages[]       # Conversation about training
├── artifacts[]      # Memory calculators, cost estimates
├── scenarios[]      # Training workload definitions
└── stacks[]         # Hardware configurations

workspace_2 (Inference Provider Comparison)
├── sources[]        # Different docs entirely
├── messages[]       # Separate conversation
├── artifacts[]      # Different analyses
├── scenarios[]      # Inference workload definitions
└── stacks[]         # Different hardware configs

When you query “what GPU should I use?”, the Research Agent only searches sources within the current workspace. No cross-contamination.

Switching Between Contexts

When you need to context-switch—an urgent request interrupts your deep work—just click the workspace selector and choose a different workspace.

The switch is instant:

• Sources panel updates to show the other workspace’s documents
• Lab panel loads that workspace’s conversation history
• Studio panel displays its artifacts and configurations

Your original workspace remains exactly as you left it.

Real-World Use Cases

Evaluating multiple model providers: Create separate workspaces for each contender: “Anthropic Claude Evaluation”, “OpenAI GPT-4 Evaluation”, “Google Gemini Evaluation”. Each contains only that provider’s documentation.

Managing concurrent projects: Maintain workspaces aligned with your project portfolio: “Project Alpha - Training Infra”, “Project Beta - Inference Pipeline”. Team members see only relevant context.

Exploring framework options: Keep “LangGraph Deep Dive” separate from “Custom Agent Architecture”. Each accumulates sources and artifacts over weeks without mixing.

What You’ve Accomplished

You now have a dedicated workspace for your AI infrastructure research. This workspace will contain all your sources, conversations, scenarios, and generated artifacts—keeping your analysis organized and accessible.

Next steps: Add sources to give Lattice context for your specific use case.

Workspace Management is available in Lattice 0.4.0. Create your first workspace and start a focused investigation today.