Roovie Workspace Agent

Roovie includes a workspace agent that connects to your choice of AI provider and operates directly on the building model.

This is not a chat window with a fixed set of commands. It is a general-purpose AI workspace where you can ask the agent to do anything the platform supports — inspect building data, size HVAC systems, create assemblies, validate models, read simulation results, make targeted mutations to building entities, and more. You bring the context by dragging in buildings, zones, assemblies, simulations, or documents. The agent works with what you give it.

The workspace connects to OpenAI (Codex) or Anthropic (Claude). You choose which provider powers the session. And when the built-in capabilities are not enough, you can write custom skills — structured markdown documents that teach the agent new workflows specific to your organization's building energy practice.

In One Line

Drag in building context. Ask the agent to do anything. Connect your own LLM. Add skills for workflows that matter to you.

How It Works

Open the workspace
  → Connect to OpenAI (Codex) or Anthropic (Claude)
  → Drag buildings, zones, assemblies, simulations, or documents into context
  → Ask the agent questions or give it tasks
  → The agent reads live building data, performs calculations, and stages changes
  → Preview mutations before they are applied
  → Attach custom skills to extend what the agent can do

The Workspace

The agent workspace is a unified environment that hosts multiple specialized agents under one interface. A switcher bar along the top lets you move between agents, each with its own conversation history, context, and visual preview panel.

Available Agents

• Roovie — the general-purpose agent. Handles open-ended questions, building inspection, cross-domain tasks, and anything that does not fit a specialized agent.
• HVAC Design — designs and sizes HVAC systems from natural language. Supports load calculation, equipment selection, and system configuration.
• Assembly Design — creates wall, roof, floor, and fenestration assemblies from descriptions or performance targets.
• Materials — generates individual building materials with thermal, surface, and environmental properties.
• Validation — checks building models against ASHRAE 90.1, 62.1, Title 24, and internal consistency rules.
• Template Library — searches, recommends, and compares templates across the material, assembly, system, and schedule libraries.
• Building Wizard — creates new buildings from scratch with guided AI assistance.

Each agent maintains its own conversation history. Switching between agents preserves context — you can start a question with the HVAC agent, switch to validation to check something, and return to HVAC without losing your place.

Context Drag-and-Drop

The workspace accepts dragged context from anywhere in Roovie's sidebar and content panels. Context types include:

• Buildings — the agent receives the building's metadata, geometry, location, and configuration
• Zones — zone dimensions, surfaces, schedules, and internal loads
• Assemblies — wall, roof, and window constructions with layer details
• Materials — thermal properties, surface characteristics, and environmental data
• Systems — HVAC equipment, capacity, zone coverage, and control strategies
• Schedules — occupancy, lighting, equipment, and thermostat profiles
• Simulations — performance summaries, monthly results, thermal metrics, energy breakdown, costs, and emissions
• Documents — parsed document content for reference
• Agent Skills — structured instructions that extend agent behavior

When context is attached, the agent has direct access to the live data — not a summary or description, but the actual building entities as they exist in the database.

Page Context

When viewing a building, simulation, or other entity in the content panel, the workspace automatically detects relevant context and offers a one-click button to send it to the agent. This means you do not have to manually drag items — the workspace knows what you are looking at and makes it available.

LLM Provider Selection

The workspace supports two AI providers:

OpenAI (Codex)

Connect to OpenAI's models for agent reasoning and tool execution. The agent bootstraps with a scoped session that includes the building context, available tools, and mutation capabilities.

Anthropic (Claude)

Connect to Anthropic's Claude for the same capabilities. The provider selection is made at session launch — both providers have access to the same building context, tools, and mutation workflow.

The provider choice is per-session. Different team members can use different providers based on preference or organizational policy.

What The Agent Can Do

Read Building Data

The agent can inspect any entity in the building model:

• Read zone dimensions, surfaces, and properties
• Read assembly layers with material thermal data
• Read HVAC system equipment and zone coverage
• Read simulation results including hourly data
• Read schedules, documents, and metadata

All reads use live data from the database, not cached snapshots.

Preview and Apply Mutations

The agent follows a structured preview-then-apply workflow for all changes:

1. The agent builds a mutation request based on the user's instructions
2. The request is staged with a preview that shows exactly what will change — which fields, which entities, what clones will be created, what relationships will be linked or unlinked
3. The user reviews the preview summary
4. Only after explicit confirmation does the agent apply the changes
5. Every applied change receives a unique action ID for traceability
6. Changes can be reverted using the action ID if needed

Supported mutation operations include:

• Update — modify fields on a single entity
• Mass update — modify fields across multiple entities
• Link — create relationships between entities (zone to system, schedule to zone)
• Unlink — remove relationships
• Assign template — apply a template with proper cloning of dependent resources
• Clone and attach — deep clone an assembly or system with all linked materials and schedules

The agent never bypasses the preview step. Every mutation is visible and reversible.

Simulation-Aware Sizing

When a simulation is attached to the context, the agent reads the simulation results directly — including hourly data with thermal, zone, and building field groups. This means HVAC sizing, load analysis, and performance review use actual simulation output, not estimates or assumptions.

The agent can read hourly thermal balance data, zone peak loads, coincident building peaks, ventilation requirements, and unmet load hours to produce sizing inputs grounded in physics.

Custom Skills

Skills are the extension mechanism that makes the workspace agent adaptable to any building energy workflow.

What A Skill Is

A skill is a structured markdown document that teaches the agent a specific workflow. It is not code — it is a set of instructions written in natural language with a defined structure that the agent follows when the skill is relevant.

Skill Structure

Every skill has eight sections:

Section	Purpose
Goal	The exact outcome the skill should produce
Use When	When the skill applies and when it does not
User Must Do	What inputs or decisions the user provides
Agent Must Do	The specific actions the agent performs
Required Context	Data sources, files, or simulations needed
Workflow	Step-by-step procedure the agent follows
Output	Expected deliverables and format
Constraints	Rules, limits, and things the agent must avoid

An optional Draft Notes section holds raw source material or ideas that inform the skill but are not part of the operational instructions.

How Skills Are Used

Skills are attached to the workspace context like any other item — drag them from the skills sidebar into the agent's context panel. Once attached, the agent incorporates the skill's instructions into its reasoning.

Multiple skills can be active simultaneously. The agent synthesizes them with the current building context and conversation.

Built-In Starter Skills

Roovie ships with starter skills that demonstrate the pattern:

Size Real HVAC From Ideal Loads Results — translates a completed Ideal Loads simulation into real HVAC sizing inputs. The skill instructs the agent to extract zone sensible peaks from hourly data, find coincident building peaks, check unmet loads at peak hours, derive airflow when supply delta-T is provided, and present results with clear assumptions and limitations.

Use Roovie Mutation Tools — keeps the agent inside the correct preview-and-apply workflow for building mutations. The skill instructs the agent to verify the session is bootstrapped, read live entities before making changes, stage every mutation as a preview, summarize changes for the user, wait for confirmation, and use revert for rollback when needed.

Writing Your Own Skills

Organizations can create skills for any building energy workflow:

• A skill that guides HVAC system selection based on climate zone and building type
• A skill that enforces a specific envelope design standard for all new assemblies
• A skill that extracts Title 24 compliance data from simulation results and formats it for submittal
• A skill that reads a parsed specification document and validates building model inputs against it
• A skill that generates ECM cost estimates using organization-specific unit costs

Skills are scoped to the organization — they are available to all team members but do not affect other organizations.

The structured format ensures skills are consistent, complete, and machine-interpretable. The validation system checks that all required sections are present before a skill is saved.

Skills as Living Documents

Skills are markdown documents stored in Roovie's document system. They can be versioned, tagged, searched, and shared like any other document. They support the full document lifecycle — create from scratch, create from a legacy prompt, edit, preview, and delete.

The Terminal Experience

The workspace uses a terminal-style interface with a monospace font, prompt prefix, and keyboard-driven input. This is intentional — the interaction model is designed for practitioners who want to work efficiently, not click through menus.

Input

• Enter sends the message
• Shift+Enter adds a new line for multi-line prompts
• The prompt auto-resizes as you type

Output

Agent responses render with full markdown formatting — headers, lists, code blocks, bold, italic, and inline code. Responses are copyable per-message and as a full transcript.

Themes

The workspace supports 10 terminal themes: Roovie Dark (default), Dracula, Solarized Dark, Monokai, Nord, One Dark, Tokyo Night, Catppuccin Mocha, Gruvbox Dark, and Cyberpunk. Themes control background, foreground, cursor, selection, and the full 16-color ANSI palette.

Visual Preview Panel

Specialized agents (HVAC, Assembly, Materials, Validation) produce visual previews alongside their text responses. The preview panel shows:

• HVAC system configurations with equipment details
• Assembly cross-sections with layer properties
• Material specifications with thermal data
• Validation issue lists with severity, auto-fix suggestions, and compliance status

The panel auto-expands when a preview is available and can be collapsed when not needed. Preview indicators on the agent switcher bar show which agents have pending previews without switching to them.

Chat History

Conversations are persisted and can be loaded later. The workspace supports:

• Conversation IDs for tracking sessions
• Message history with role (user or assistant), content, and timestamps
• LLM provider and model metadata per conversation
• History loading from the sidebar for returning to previous sessions

What Makes This Different

Most AI features in building tools are fixed chatbots — they answer questions from a predetermined set of capabilities, and when they cannot help, the user is on their own.

Roovie's workspace agent is different:

• Provider choice — connect to OpenAI or Anthropic based on your preference, not a vendor lock-in
• Live building context — the agent reads actual database entities, not summaries or exports
• Preview-before-apply — every mutation is staged, visible, and reversible
• Simulation-aware — the agent reads hourly simulation results for physics-grounded analysis
• Custom skills — organizations extend the agent's capabilities with structured markdown, not code
• Multi-agent switching — specialized agents for HVAC, assemblies, materials, and validation share the same workspace
• Terminal interaction model — keyboard-driven, markdown-formatted, copy-friendly output designed for practitioners

The workspace is not a feature you visit occasionally. It is the interface for working with AI across the entire platform — from quick questions to complex multi-step engineering workflows.

Bottom Line

The Roovie Workspace Agent is a general-purpose AI environment that connects to your choice of LLM provider and operates directly on your building data. Drag in context, ask questions, stage mutations with preview and confirm, and extend the agent with custom skills for any building energy workflow.

The combination of live building context, structured mutation tooling, simulation-aware reasoning, and user-defined skills makes it an extensible engineering assistant — not a fixed chatbot, but a workspace that grows with how your organization works.