The Roovie AI Layer

AI at the edge. Physics at the core.

Agents that generate HVAC, assemblies, and materials. Agents that build whole models. Agents that run simulations and read results. Agents that read every document you've uploaded. A Copilot that sits beside you and already knows what you're working on. AI doesn't replace the engineer — it erases the manual work around them.

HVAC agentAssembly agentMaterials agentValidation agentBuilding wizardSimulation insightsCopilot · context-aware

Put an agent to work See why it works

▸ Why Roovie's AI is different

Agents that live inside the system — not bolted on top of it.

Most AI features in building software are chat widgets duct-taped to a legacy file format. Roovie owns the physics engine, the data model, and the object graph end-to-end. Which means our agents speak the same language as the app itself — every zone, surface, assembly, material, simulation, and document is addressable, queryable, and mutatable.

The bolt-on pattern

LLM wrapped around a legacy tool

×Chat lives outside the model. It can describe the building but can't touch it — or touches it through a fragile export/import round-trip.
×Every agent re-learns the schema. No shared object graph, so each feature reinvents how to 'find a zone' or 'apply an assembly.'
×No context portability. Jumping from HVAC to assemblies to simulations means starting a new conversation, re-attaching the same files.
×No safe mutations. Edits either bypass validation entirely or require a manual reload of the whole file.
×No simulation native to the agent. To check an idea, the AI writes a recommendation — you run the simulation yourself.

The Roovie pattern

Agents native to the platform

✓Single object graph. Zones, surfaces, assemblies, materials, systems, schedules, simulations — one addressable model every agent shares.
✓Agents jump across the app. HVAC → Assembly → Materials → Validation inherits the same context. No re-attaching. No re-explaining.
✓Preview, apply, revert. Every mutation is staged for human review and reversible with one command. Safety by architecture, not policy.
✓Simulation is a first-class tool. Agents trigger real runs on our own engine, read hourly results, and reason about them directly.
✓Custom materials and assemblies understood natively. The agent reads your library the same way the physics engine does — no translation layer, no loss.

▸ The agent bar

Switch agents the way you switch tabs. Context follows.

Every workbench carries the same top-level agent bar. Pick the agent for the job — HVAC, Assembly, Materials, Validation, Templates, Building Wizard — and it inherits the active building, the selected zone, the attached simulation. Nothing to re-upload. Nothing to re-explain.

HVAC· context: Dallas Tower A · zone Z-101

Give this building a VAV-reheat system. Gas heat, electric cool, 0.8 COP reheat target.

Sizing from the attached Ideal Loads run. Coincident building sensible peak is 168 kW (Aug 14, 3 p.m.). Splitting into 4 VAV zones matched to perimeter groups + 1 core terminal. Gas boiler for reheat and DHW, electric DX for cooling.

read buildingread simulationcalculate loadspreview VAV system

Preview looks right. Apply it.

Ask the HVAC agent anything...Send

▤ Preview · VAV system, sized to the modelplan_7a2f

revertapply · plan_7a2f →

▸ Generate what you'd otherwise build by hand

Describe it. Roovie builds it.

HVAC systems, wall and roof assemblies, custom materials — the parts that used to mean clicking through ten forms or hunting a library for half an hour. Describe what you want in plain language and the right agent generates it with real properties, real schedules, and real ties to the model.

HVAC agent

system sizing · equipment · controls

Prompt"Give the whole building a VRF system with 4-pipe DOAS. Cooling COP 4.5, efficient range."

Assembly agent

wall · roof · floor construction

Prompt"Build me an R-30 mass wall. Exterior brick, 6 in CMU, continuous insulation, 5/8 gyp."

Materials agent

custom materials · thermal · optical

Prompt"Make a phase-change gypsum material. 45 °C transition. Match our spec sheet attached."

▸ Autonomy · end-to-end

Generate the model. Run the simulation. Read the results.

Every node of the building-to-insight pipeline has an agent that can drive it on its own. Hand the chain to Roovie and it runs the whole loop: model → simulation → analysis → report. Stop it at any step to review. Resume when you're ready.

01 · Building wizard●generated

Generate the building

Hand the wizard an address, a program, a floor count, and a vintage. It returns the footprint, stacks the floors, applies core + perimeter zoning, and drops in assemblies from your org library.

02 · Simulation runner●10 in parallel

Run baseline + ECMs

The agent queues a baseline, then fans out ECM variants — envelope, HVAC, lighting, controls — and runs them concurrently on the cloud. No IDF export. No file-staging. Just results streaming back.

03 · Simulation insights●analyzing

Read the results · draft the report

The insights agent reads hourly outputs, spots the top drivers, compares scenarios, flags unmet loads, and drafts charts + narrative. You edit what matters, ship the rest.

↳Every step previews before it applies. Every step is reversible. Every step leaves an audit trail — so autonomy never means you've lost track of what the model just did.

▸ The Copilot · always in the room

Ask anything. It already knows what you're looking at.

The Copilot is the general-purpose agent that lives at the edge of every workbench. Whatever building you have open, whatever zone is selected, whatever simulation is loaded, whatever report is on your screen — it shares that context automatically. No uploading. No re-pasting. Just ask.

▤ copilot · Dallas Tower A
context · 3 items attached

Attached · from what you have openbuilding Dallas Tower A · 8 floors · 38 zones
zone Z-101 Perim NW · selected in viewport
simulation sim_2a88 · ECM · R-25 walls (just finished)
› Why did the cooling load in Z-101 jump 18% between 2 and 4 p.m. in the ECM run?
Reading the attached hourly results. Solar gains through the north-orientation glazing spike at 15:00 on Aug 14 — the low sun reaches the interior because your WWR on that façade is set to 50%. The ECM's lower wall U-value also stores less heat to buffer the peak.
Z-101 · cooling load · Aug 14baseline vs ECM
14–16 H SPIKE0612151824 h+18%baselineECM · R-25 walls
Driver split for the spike: solar 64 % internal 22 % vent 14 %
Quick actions the Copilot can take from here→ Rerun this ECM with WWR N = 30%
→ Generate an interior shade schedule from the insights agent
→ Draft a one-page memo on the tradeoff for the design team
› Rerun it with WWR North at 30% and summarize the delta

The Copilot sees what you see

▸Active building and viewport. The 3D scene, the zone you selected, the surface you clicked.
▸Running and finished simulations. Including hourly results — no download, no re-attach.
▸Open documents. Any spec, report, or memo that's currently loaded in a tab.
▸Your materials and assemblies library. Org-scoped, already shaped the way your team uses it.
▸What you just did. The last edits, the last runs, the last plan it applied — all in the audit trail.

What it can do

▸Answer questions about the app — where to find something, how a feature works, which agent does what.
▸Analyze on demand. Forward the current context to any specialized agent and summarize the result.
▸Drive the other agents. Orchestrate HVAC → Assembly → Validation without losing the thread.
▸Preview every change. Same safety rail as the specialized agents — nothing writes without your OK.

▸ Talk to the pile of PDFs no one reads

Every document you upload becomes searchable knowledge.

Specifications, reports, contracts, product cutsheets, code documents, vendor O&M manuals — drop them in, and Roovie indexes them into the same knowledge layer the agents already use. Ask a question, get an answer with citations back to the exact passage.

Indexed · 48 documents

PDF

Owner design spec · v3.2

142 pages · 4 days ago

PDF

ASHRAE 90.1-2022 excerpt

88 pages · 2 w ago

DOCX

Vendor O&M · Carrier 30XA

36 pages · 1 w ago

PDF

PCM-Gyp 45C spec sheet

12 pages · today

PPTX

Energy roadmap · Q2 review

41 slides · 3 d ago

Internal HVAC playbook

6 pages · 5 d ago

XLSX

Portfolio 2026 budget

11 sheets · 2 w ago

→ Dropped in, converted, chunked, and embedded in the same pipeline the agents use. Nothing lives outside the system.

Does the owner spec allow us to substitute a VRF system for the originally specified chiller loop? And what efficiency tier do they require?

Yes — with conditions. Section 23 81 26 allows VRF as an "equal alternate" provided it meets AHRI-rated IEER of at least 21.5 at full load and 27.1 at part load, and includes factory-installed branch controllers. The cooling COP floor in the same section is 4.1 — the VRF system the HVAC agent generated (COP 4.6) clears that by a comfortable margin.

And what does 90.1 say about our proposed envelope for climate zone 3A?

For CZ 3A mass wall, the prescriptive U-factor ceiling is 0.090 W/m²K. Your assembly (U = 0.033) is 63 % below the prescriptive cap — plenty of room on the performance path as well.

→ same thread, same context, same building · agents and docs share the room

▸ Extend · skills and agents of your own

Write your own skills. Ship your own agents.

Skills are short, markdown-style playbooks that teach any Roovie agent how your team does its work — the sizing method you prefer, the benchmarks you hit, the validation rules you enforce. Agents are full new personalities you can build on top of the same object graph: a physics agent, a weather modeling agent, a building-validation agent specific to your portfolio.

▤ size-real-hvac-from-ideal-loads.mdscope · organization

# Size Real HVAC From Ideal Loads Results

## Goal
· Translate a completed Ideal Loads simulation into real HVAC sizing inputs.
· Produce zone sensible peaks, coincident building peaks, ventilation requirements.

## Use when
· A building has an Ideal Loads run with hourly data.
· The user is moving toward a real HVAC system.

## Agent must
· Size each zone sensible peak from zone_cooling_load, zone_heating_load.
· Size central equipment from coincident loads, never summed peaks.
· Separate sensible, latent, and ventilation.
· Flag unreliable hours from unmet loads.

## Tags

hvacsizingideal-loadssimulation

✓ attached to HVAC agent · Copilot · Simulation insights

Custom agents · ship them like skills

Start from a built-in agent or author one from scratch. Define its tools, scope, skills, and the model objects it's allowed to touch — same preview / apply / revert rails, same audit trail.

Physics agent

Explains results, suggests fidelity changes, cross-checks heat balance closure per timestep.

yours · org

Weather modeling agent

Pulls climate data, builds TMY / AMY overlays, compares future-weather scenarios side-by-side.

yours · user

Portfolio validation agent

Enforces your team's modeling standards across every building: zoning rules, required assemblies, mandatory schedules.

yours · org

+ Draft a new agent · scope, skills, and tools in markdown

▸ The payoff

AI isn't the center. It's the accelerator.

The physics engine still runs the physics. The engineer still owns the decisions. The AI agents erase the manual work between them — clicking through forms, hunting for materials, staging simulations, rummaging through PDFs. Months of takeoff collapse into hours. Hours of analysis collapse into minutes. The twin stays live. The team stays in the loop.

Manual modeling · analysis · review

days → weeks

Native agents · shared graph

With Roovie · AI-accelerated

Minutes · live · reviewable

Client Portal Coming SoonContact

Generate HVAC · assemblies · materials●Autonomous simulation●Chat with your docs●Build your own agents