Nohm ONE: AI Optimisation Networks for Energy Planning | Full Webinar | Terajoule

Executive Summary

Terajoule presented Nohm ONE as an AI-enabled orchestration and analysis layer for energy planning—not a replacement for established modeling tools. It aims to connect data, models, workflows, and analysis so planning teams can reuse prior work, explore scenarios more quickly, trace results to their inputs, and make complex studies easier to review.

The webinar was a product preview. Some capabilities were demonstrated live, while more autonomous and “shape-shifting” modeling features remain under development.

Business Problem and Target Users

The platform is intended to address four recurring problems in energy planning:

• Fragmented information: Tools, files, and assumptions are spread across systems, and changes may not propagate through the full workflow.
• Repeated preparation: Planning cycles often restart from scratch, with limited reuse of prior models and analysis.
• Dependence on scarce specialists: Non-modelers rely on busy energy modelers to interpret model inputs and results.
• Difficult review: Increasingly granular, interconnected, multi-carrier models generate results that are hard to inspect and explain.

Target users include system-planning teams, project promoters conducting cost-benefit analysis, and analytical organizations such as consultancies, think tanks, and institutions.

Product and Operating Approach

Nohm is designed as a tool-agnostic orchestration layer. It uses a shared internal model schema and can connect to existing modeling, demand, climate, and other tools through APIs or MCP servers. Tools mentioned included PLEXOS, PyPSA, ETM, and bespoke Python scripts. The goal is to coordinate existing systems rather than displace them.

The platform’s intended benefits include:

• Reusing and referencing prior studies, assumptions, and model versions.
• Tracing results back to their sources, including when investigating unexpected behavior or model issues.
• Making model knowledge more accessible to teams beyond specialist modelers.
• Supporting review through analysis and visualization, rather than requiring users to inspect raw data alone.

Demonstrated Product Areas

• Agent-based assistance: A coordinating assistant routes questions to specialist agents, including those for energy modeling, economics, communications and visualization, and IT-related functions. Users can ask questions, request analysis, or initiate model changes.
• Model and data management: Workspaces hold databases, projects, models, analysis history, and version information. Model changes are recorded so teams can inspect how a model evolved.
• Analysis: “Shallow” analysis answers focused questions, while “deep” analysis decomposes a question into contributing data and time steps. Comparative analysis can identify differences between model runs and present relevant charts.
• Atlas model exploration: Users can inspect network structure, model parameters, flows, and results; compare runs; and create model subsets at different geographic resolutions.
• Visualization: Results can be viewed as maps, charts, and time series. A Grid Flow Explorer shows flows over time and highlights lines approaching capacity; the demo used a 99% capacity threshold as an example.
• Connected data: The proposed database combines TYNDP scenarios with national development plans and market intelligence, with the goal of keeping supply, demand, grid, and storage data up to date and available at finer spatial resolution.
• Cost-benefit analysis: The demo showed how project scenarios can be compared for effects on system costs, producer rents, consumer prices, and congestion.

Frameworks, Processes, and Playbooks

• Client delivery sequence: Discovery and scoping → proof of concept → full implementation once the proof of concept is approved → optional research and innovation work.
• Model exploration workflow: Start with a system-level model → zoom into a problem area by changing geographic granularity → investigate potential mitigations → compare results to inform a decision.
• Proposed AI feedback loop: Define research questions and KPIs → run the model → analyze model behavior and unexpected results → propose and validate changes → repeat until the gap to the target is acceptable. Engineer oversight remains part of the process.
• Proposed “shape-shifting” model approach: Adapt a model to the question by changing its carriers, demand-sector detail, geographic clustering, time resolution, or climate assumptions; repair or patch links affected by the changes.
• Deployment choices discussed: Use external AI services; use API-based services that do not retain or train on company data; host models on a virtual private server; or run models on the organization’s own infrastructure.

Concrete Examples

• Mixed-resolution modeling: In the Atlas demo, a focus country could be modeled in greater detail, neighboring countries at an intermediate level, and the rest of the system at bidding-zone level. Users could also regroup countries into regions and build a model from the selected topology.
• Model comparison: The platform compared flows and prices across weather-year variants, using colors to indicate increases or decreases and an AI-generated interpretation of whether a change was favorable.
• Lunar-base study: Terajoule showed an experimental four-person lunar-base model linking electricity, solar and nuclear generation, storage, water, oxygen, hydrogen, electrolysis, fuel cells, rovers, and a greenhouse. It was presented as an illustrative multi-energy-system example, not a commissioned project.
• Model subset creation: A live demonstration created a model subset for Spain and France and discussed whether to preserve or omit cross-border flows. The platform’s internal schema is intended to support running the resulting model in an external engine such as PLEXOS or PyPSA.

Metrics, Targets, and Timelines

• Terajoule described its team as having about five staff members.
• The presenter reported more than 10 years of energy-planning experience.
• The platform was described as approximately 60–70% of the way through “stage two” of development. The stated next stages are shape-shifting models, followed by a more autonomous “living architecture” in which models interact and coordinate studies—with engineering oversight.
• Terajoule hoped to offer an initial preview by the end of 2026, followed by a more comprehensive version the following year. The commercial model and hosting and security arrangements were still being worked out.
• The platform supports analysis at granularities including hourly and, in some contexts, 15-minute data. Model-resolution examples included 220 kV network detail.
• No revenue, margin, customer acquisition, retention, or other commercial performance KPIs were provided.

Commercial and Product Caveats

Nohm was described as a preview, and the presenter noted that agents were still being trained and could produce inconsistent results. A live query did not return a satisfactory answer; the presenter attributed this partly to the system’s nondeterministic behavior and preview status.

PLEXOS use currently requires the customer to have a PLEXOS license, with the demonstrated integration using its API. The platform was not open source at the time of the webinar. Terajoule was considering access options while developing its service, security, and infrastructure model.

Presenter and Sources

• Presenter: Dante Powell, Terajoule.
• Sources and tools referenced: Terajoule’s Nohm platform and demonstrations; TYNDP scenarios; national development plans and market intelligence; PLEXOS, PyPSA, and ETM.