What powers Sympheny — and what doesn't.
A clear explanation of how Sympheny calculates, what the clustering does and does not do, and where AI assistants fit — plus precisely what happens to your data.
Three distinct layers — each transparent and purposeful.
Sympheny operates across three layers. The first two run inside the calculation: the optimiser itself, and the clustering that makes a full year of hourly data solvable. The third sits outside it — the interfaces you can drive Sympheny through, including AI assistants you choose to connect. Nothing in the third layer changes what the first two compute.
The core of Sympheny is a Mixed-Integer Linear Programming (MILP) engine — a rigorous mathematical approach used in engineering and operations research. It finds the optimal energy system configuration given your technical, economic, and regulatory constraints.
Every result is fully deterministic and auditable: the same inputs always produce the same outputs, and every recommendation can be traced back to its underlying equations. There is no black box, and no AI in the calculations.
Solving a full year of hourly data directly would be impractical, so Sympheny groups similar days into a smaller set of representative typical days using unsupervised clustering. You control how many through the temporal resolution setting, which also fixes the error tolerances the clustering has to meet.
This is grouping, not learning. No model is trained, nothing is carried between runs, and your data is never used to improve anything. The clustering is computed fresh from your own inputs every time, and re-running an unchanged scenario produces the same typical days and the same results.
Sympheny publishes a REST API, a Python SDK and an MCP server, so an AI assistant or coding agent can set a scenario up, start an optimisation run and read the results back. Setup is documented at docs.sympheny.com. This is an interface to the platform, not a change to how it calculates: an assistant can build the model, but the model is still solved by the MILP engine in Layer 1, and the same inputs still produce the same outputs.
None of this is on by default. You decide whether to connect an assistant and which provider to use, and the data you send reaches that provider under your agreement with them, not ours. Sympheny embeds no third-party model in its own computation pipeline.
Sympheny is built on a mathematically rigorous MILP core, supported by unsupervised clustering that groups your data without ever learning from it — reproducible run to run, and with no model trained on customer data.
Need procurement documentation?
We can provide a Data Processing Agreement, sub-processor list, and a technical description of the platform's AI components on request. Talk to us before procurement — we make the review easy.