AI-Powered Design Agents

Roovie includes seven specialized AI agents that operate inside the building workbench.

Each agent is trained on a specific domain of building design and engineering. Users interact through natural language — describing what they need rather than navigating menus or filling forms. The agents parse intent, gather building context, perform calculations, and produce previews that users confirm before anything is saved.

This is not a chatbot bolted onto a building tool. Each agent has access to Roovie's building model, calculation engines, material libraries, and validation systems. They do real engineering work through structured tool-calling workflows.

In One Line

Describe what you need. The agent designs it, sizes it, validates it, and previews it for your confirmation.

The Seven Agents

1. HVAC Design Agent

Designs complete HVAC systems from natural language descriptions.

The agent supports three workflows:

• Natural language design: "I need a VRF system for a 3-story office in Climate Zone 4A" — the agent parses the request, calculates loads, selects equipment, sizes capacities, configures controls, and builds the complete system.
• Known system input: "We have two 15-ton Carrier rooftop units with gas heat" — the agent extracts specifications from the description, validates against standards, calculates any missing values, and produces a system preview.
• Interactive design: The agent asks clarifying questions about sizing method, system type, fuel source, and efficiency targets, then auto-generates a system based on the conversation.

The HVAC agent supports four load calculation methods:

• Rule of thumb: Quick estimation based on building type and area
• Envelope-based: Calculation from building geometry, assemblies, and climate
• User input: Manual load values provided by the engineer
• Simulation-derived: Loads extracted from a completed physics engine run

It handles 19+ HVAC system subtypes including packaged rooftop units, split systems, VAV, DOAS, fan coils, VRF, PTAC/PTHP, chiller and boiler plants, radiant systems, and ground-source heat pumps.

The agent also manages thermostat schedules with heating and cooling setpoint profiles, and can verify completed designs against ASHRAE 90.1 and 62.1 standards.

2. Assembly Design Agent

Creates building envelope assemblies — walls, roofs, floors, and fenestration — from descriptions or performance targets.

• "Create an R-20 metal roof assembly" — the agent selects appropriate layers (exterior finish, insulation, structure, interior finish), chooses materials, calculates thickness to meet the R-value target, and previews the complete assembly.
• "Analyze the current wall assembly and suggest improvements" — the agent inspects the existing assembly, identifies thermal weaknesses, and recommends specific layer changes.

The agent understands layer zones (exterior finish, sheathing, weather barrier, insulation, vapor barrier, structure, interior finish) and can search existing template libraries or generate new materials with calculated thermal properties.

3. Materials Agent

Creates and validates individual building materials.

• "Create R-30 fiberglass batt insulation" — the agent generates the material with correct thermal conductivity, specific heat, density, and R-value.
• "Make it more sustainable" — the agent refines the material to a lower embodied carbon alternative while preserving thermal performance.

All materials are validated against ASHRAE reference data. The agent calculates derived properties (thickness from R-value, or R-value from thickness and conductivity) and supports the full property set: thermal, surface, hygrothermal, fire, environmental, and cost.

4. Building Validation Agent

Performs comprehensive validation of a building model before simulation.

The agent checks across 10 categories:

• Geometry: zone dimensions, vertices, volume calculations
• Surfaces: surface accounting, window-to-wall ratio, opening placement
• Assemblies: assignment completeness, layer properties, R-value verification
• Materials: property bounds, consistency
• Internal loads: lighting, equipment, and occupancy density reasonableness
• HVAC sizing: capacity versus loads, zone coverage, schedule assignment
• Schedules: assignment completeness, setpoint reasonableness
• Title 24: California-specific compliance checks
• Ventilation: ASHRAE 62.1 requirements
• Building consistency: cross-component validation

Each issue is categorized by severity (error, warning, info) and includes a unique code, detailed message, and — where available — auto-fix suggestions with confidence scores, impact levels, and reversibility indicators.

The agent can also perform root-cause analysis using LLM reasoning, identifying fix ordering and quick wins across the full issue set.

5. Simulation Insights Agent

Analyzes simulation results and generates insights.

Users drag and drop simulation variables into the agent's workspace. The agent analyzes the data and produces:

• Observations: factual findings about the data patterns
• Recommendations: actionable suggestions based on the analysis
• Warnings: potential issues or anomalies detected
• Comparisons: how results relate to benchmarks or other simulations

Each insight includes confidence level, impact rating, and links to the relevant variables. The agent can also generate chart specifications and save visualizations as documents.

6. Template Library Agent

Searches, recommends, and compares templates across the full library.

• "Find a high-efficiency wall assembly for Climate Zone 5" — the agent searches materials, assemblies, HVAC systems, and schedules using a scoring algorithm that weighs type match, R-value proximity, keyword relevance, usage popularity, and verification status.
• "Compare these three wall assemblies" — the agent produces a side-by-side matrix with key metrics highlighted.

The agent parses search, recommend, compare, and detail intents automatically from natural language input.

7. Roovie General Agent

A context-aware assistant that maintains a session tied to the current building and selected components.

Users can focus the agent on specific zones, assemblies, systems, materials, schedules, or simulations. The agent maintains conversation history and can answer questions, explain building behavior, and coordinate with the specialized agents when a request crosses domains.

How Agents Work

All agents follow a consistent interaction pattern:

User describes what they need (natural language)
  → Agent parses intent and gathers building context
  → Agent asks clarifying questions if needed
  → Agent performs calculations and generates a preview
  → Preview is displayed in the visual panel with full details
  → User confirms, refines, or rejects
  → On confirmation, the result is saved to the building model

Nothing is saved until the user confirms. Every agent produces a preview first.

Agentic Tool Calling

The HVAC Design Agent supports a V2 agentic mode where the AI autonomously selects and calls tools to accomplish the design task. Available tool categories include:

• Information tools: query building data, zone info, climate zones
• Calculation tools: load calculations, ventilation requirements, sizing analysis
• Equipment tools: equipment suggestions, efficiency calculations
• System tools: system configuration, verification, template application

The agent explains its reasoning as it works, making the design process transparent and auditable.

Streaming Responses

The General Agent supports server-sent event streaming, providing real-time token-by-token responses for conversational interactions.

Organization-Level Customization

Agent prompt templates can be customized at the organization level:

• System templates provide the default behavior for all users
• Organizations can fork system templates and customize them
• Protected sections within prompts cannot be modified, preserving physics compatibility
• Variable support allows templates to reference building-specific data dynamically
• Templates are versioned and can be set to draft, active, or deprecated status

This means organizations can tailor agent behavior to their specific standards, terminology, and design preferences without breaking the underlying engineering logic.

What Makes This Different

Most "AI in building design" features are chatbots that answer questions about buildings. They can discuss concepts but cannot do the work.

Roovie's agents are different:

• They have direct access to the building model, not just a description of it
• They perform real calculations using the same engines as manual workflows
• They produce structured outputs (HVAC systems, assemblies, materials) that become part of the building
• They validate against real standards (ASHRAE 90.1, 62.1, Title 24)
• They follow a preview-then-confirm pattern that keeps the user in control
• They can be customized per organization without losing engineering rigor

The agents do not replace engineering judgment. They accelerate it by handling the mechanical parts of design — searching libraries, sizing equipment, calculating properties, validating compliance — so engineers can focus on the decisions that matter.

Bottom Line

Roovie's AI agents turn natural language into engineering artifacts. They design HVAC systems, create envelope assemblies, generate materials, validate buildings, analyze simulations, and search libraries — all through conversation.

The key is that they operate inside the building model, not alongside it. What they produce is real, structured, validated building data that feeds directly into Roovie's simulation, visualization, and reporting workflows.