Your consumption data, fetched from where it already sits
An energy monitoring project almost always fails in the same place: the data. It sits with the distributor, on an FTP server, inside an invoice PDF, or in a spreadsheet somebody fills in by hand. Softee goes and gets it from those four places, and no Softee hardware has to be installed to do it.
How many data sources, and are they active by default?
Eleven data sources are connected out of the box, and their default state is enabled: the function that decides activation returns true in the absence of configuration to the contrary. Each can be disabled individually through deployment configuration.
Each source carries a daily scheduled job. The same scheduler then chains the calculation pipeline, the summary snapshots and the alert evaluation — in that order, otherwise you would be evaluating yesterday’s data. The day’s data is therefore collected, calculated and checked without anyone asking for it.
How does Softee fetch remote reads from the distributors?
Through three distinct connectors, written against three different interfaces. These are infrastructure operators: naming them is factual and implies no commercial partnership.
Electricity — Enedis
Collects remote-read files dropped on SFTP, decrypts them with AES-256 — the plaintext is a ZIP archive — then normalises them. Six flow types recognised: R50, R63, R64, R151, R171, R4X. Processed files are moved to a dedicated folder and failures are reported by email.
Gas — GRDF
Calls the ADICT v2 API over OAuth2, with a retry policy on error, NDJSON parsing and access-right handling supply point by supply point.
Climate — Météo-France
Orders a climatological station’s file, downloads it, and computes heating degree days on an 18 °C base, cooling degree days on a 26 °C base and minimum, maximum and mean temperatures. The station used is the nearest active station to the site, resolved from its latitude and longitude — or one you pin by hand.
What if my data does not come from the distributor?
The other eight sources cover the ordinary paths of a real estate: files dropped on FTP — several column layouts recognised, values per supply point, including water and down to minute granularity —, a token-authenticated supplier API whose account password is stored encrypted and decrypted at point of use, and index messages consumed in batches from an MQTT / AMQP queue, written idempotently on the supply point / date pair.
A separate companion service, also shipped by Softee, collects PEGAZ gas remote-read files and HELIOS HM169 concentrator files (wireless M-Bus) dropped on FTP/FTPS, normalises them into index points per supply point and publishes them to Softee over a message queue. The loop is verifiable from both ends: the producer sits in the companion service, the consumer in the product’s API.
How do you set up an estate without spending a month on it?
With a workbook. You download an .xlsx template, fill it in, have it validated as a dry run — an error report is produced before anything is written — then import it to create the sites, meters and sensors. The same prior check applies to connector imports: nothing is written on the strength of an unverified file.
Manual entry is then a data source in its own right, on the same footing as a remotely read meter rather than a fallback: a sensor’s or a meter’s data can be entered or imported for a year, and exported back out. Past years’ consumption and invoices are brought in the same way, by annual import or through the onboarding workbook.
What makes the collected data usable?
A daily check against five anomaly rules: spike, reset to zero, interpolated gap, unfilled gap, suspected meter change. A health indicator is recomputed per sensor from those checks. Softee detects and logs these anomalies.
At data level, aggregation, cumulative behaviour and missing-value handling are set indicator by indicator: a cumulative index and an instantaneous power reading do not sum the same way, and the product does not pretend otherwise. Heavy calculations go through a message bus, which allows a calculation to be re-run in a targeted way without blocking the interface.
What this page does not claim
- No energy supplier or third-party vendor named. Some sources and some invoice profiles are written against named suppliers’ interfaces. We publish the count and the technical nature, not the brands: naming an energy supplier would suggest a partnership that does not exist.
- No invoice extraction by artificial intelligence. PDF extraction in the shipped product is deterministic text extraction by library. An extractor built on a language model exists in a repository that is not deployed and is only wired up if configuration is supplied — so it does not count, and we do not announce it for later either.
- No universal coverage promised. Eleven sources connected out of the box does not mean your meters are among them. That is the first thing to check together, with the list of your supply points.
Frequently asked questions about data collection
Tell us where your data is
Which distributors, which concentrators, which suppliers, which files. We will tell you which source already exists and which would need a mapping — that is the question that decides feasibility, and it comes before everything else.