Intelligence Layer
Noda's Intelligence Layer is a set of AI agents that work through flagged faults the way an engineer would, so building teams receive a short list of Projects ready to accept.
The Project Generator groups a building's recent Insights into draft Projects that share a common fix. The Project Refiner then ranks the drafts and drops any that duplicate a Project already open.
The Root Cause Analyzer treats each Insight as evidence to test. It confirms the equipment was running, checks the behavior against multiple points and the weather, and rejects any draft the evidence doesn't support.
Faults frequently show up downstream of their cause. The Root Cause Analyzer traces symptoms through connected equipment to the source, such as a chiller bypass valve behind warm zones two systems away, and groups the symptoms under one Project.
The Cost Savings Estimator writes and runs analysis code against measured trend data, such as pump speeds and bypass flow, then adjusts the annual savings for operating schedules and seasonal weather. Each estimate carries a record of its assumptions.
flow = trend("CH-2 bypass flow", days=14) pump = trend("CHWP-2 speed", days=14) excess_kwh = plant_kwh(flow, pump) - plant_kwh_at_design_dt() annual = annualize(excess_kwh, schedule, weather)
Project Assist regularly rechecks open Projects against recent building data and marks whether each issue is still active, with a chart and a plain-language explanation as evidence. When a Project stalls, Project Assist drafts a follow-up, reviewed before it reaches the building team.
Still active. AHU-5 supplies air above its 55°F setpoint with the cooling valve fully open.
Noda's agents work from the platform's standardized data model, reasoning across equipment and systems to provide intelligent conclusions with the supporting evidence attached.
The data model links each terminal unit to its air handler and each air handler to its plant, so the Root Cause Analyzer can trace a fault upstream.
The Project Generator starts from a building's recent Insights, found by fault detection rules running across the data model.
Run commands, run statuses and static pressures confirm a unit was running. Temperatures, fan speeds and valve positions show how each unit was behaving.
Outside air temperature puts equipment behavior in context and shapes the seasonal adjustment in each savings estimate.
Occupancy schedules separate normal running from out-of-hours running and feed the annual savings.
The building's utility rates turn energy savings into dollars. Where no rate is on file, the estimate records the default rate it used.
Fault detection rules check a building's equipment data on a schedule and flag behavior outside its expected range. Noda calls each finding an Insight. Turning a list of Insights into work has typically been a manual exercise for engineering teams: grouping faults that share a cause, ruling out faulty sensor readings, tracing faults to their source and calculating what each fix would save.
In Noda's Intelligence Layer, AI agents run the triage, root cause analysis and savings calculations. The AI Project Generator regularly runs four agents in sequence over each building's recent Insights and the trend data behind them, and delivers up to 3 Projects at a time for the building team to accept. Project Assist reviews open Projects on the same cycle and checks whether each issue still shows in the data.
Noda's Root Cause Analyzer treats each Insight as evidence to test. Before a Project is created, the agent confirms the equipment was running at the time, checks the behavior against several sensors so one faulty reading can't decide the result, and reads the weather for context. A draft the data doesn't support is rejected and never reaches the building team.
Projects created by the AI Project Generator appear as created by Noda AI and carry the sparkle icon, and a filter in the Projects list shows AI-generated Projects on their own. Each Project sets out the diagnosed root cause, the Tasks to resolve it, the estimated annual savings with the assumptions behind the figure, and links to the Insights the agents worked from.
On open Projects, Project Assist's Active or Inactive badge shows when the AI last checked the data, and the Project details hold a plain-language explanation of that check and a supporting chart.
Each savings estimate is a planning figure made before work starts, grounded in measured trend data wherever that data exists. The Cost Savings Estimator works from trend data first, then equipment and manufacturer data, and uses engineering defaults only where neither is available, recording every default as an assumption on the estimate.
The agent writes the calculation as code, runs it, checks the result and reruns any figure that isn't reasonable, then adjusts the annual savings for operating schedules and seasonal weather. Estimates grow more specific to each building as Noda collects more of its data.
No. The AI Project Generator creates at most 3 Projects per building in each run. Drafts that duplicate an open Project are dropped, a Project someone has declined is never recreated, and a draft becomes a Project only when its estimated savings clear a minimum, $1,000 by default and adjustable for each building.
The building team. Each new Project arrives at Needs Acceptance with its root cause, Tasks and estimated savings, and the building team decides whether to accept it.
No. The Intelligence Layer's agents read building data and write Projects, Tasks and draft comments inside Noda. Building teams make the changes to setpoints, schedules and equipment.
Yes. The AI agents have no direct access to the building management system: they read the data Noda collects through Noda's MCP server, which gives each agent scoped access to only the data its task requires. A Noda LaunchPad at each site reads the building management system and sends data out over an outbound-only connection, so the building network stays closed to inbound traffic. Noda is ISO 27001 certified, and data is encrypted in transit with TLS 1.3 and at rest with AWS KMS.
Project Assist regularly rechecks open Projects against recent building data. The agent marks whether each issue is still Active or now Inactive and writes a plain-language explanation with a supporting chart. When an active Project has gone quiet, Project Assist drafts a follow-up comment, which is reviewed before it reaches the building team.
No. Noda adds engineering capacity to the teams already in place. The AI agents run the analysis that comes before a decision, including root cause analysis and savings estimates, and engineers decide which Projects to accept and carry out the work. As a result, the same team can cover more buildings, with more of each engineer's week spent completing fixes that already carry a savings estimate.
Noda chooses the frontier model best suited to each agent's task and isn't tied to a single AI provider. Each agent works from only the context its step requires, which keeps the analysis repeatable and the cost low enough to run across whole portfolios on a schedule.
@Craig N