How is The Timeseries Refinery different from InfluxDB or TimescaleDB?
InfluxDB and TimescaleDB are built for fast ingestion of points, and every query requires SQL, Flux or InfluxQL. What they keep is the current state of each series — past states are not reconstructible, unless you design and maintain your own versioning schema on top. The Timeseries Refinery is bitemporal by default: every insert creates a new version of the series, all versions stay queryable, and this requires zero modelling effort — it is how the store works. It also adds a traceable formula engine on top, usable by analysts entirely through the UI. It is a complete platform rather than a storage engine.
How is it different from Databricks or Dataiku?
Databricks and Dataiku are general-purpose data platforms: powerful, but they require a data engineering team to model and expose time series data before analysts can work — and bitemporal versioning would be one more thing to design and maintain there. The Timeseries Refinery is specialised: a named series catalog, bitemporal versioning by default and formulas are built in, and it is operational in minutes (SaaS) or hours (self-hosted) — no data engineering team required.
Do analysts need to write code or know SQL?
No. Formulas, series management and dashboards are fully accessible through the web interface — nothing requires code. Analysts who do code will find the full power of the platform in the Python API.
Is there a real Python API?
Yes, and it is a pillar of the system — the whole platform is built on it. Everything the web interface does goes through the same API, series come and go as pandas objects, and its design is tuned both for interactive analysis (notebooks, models, backtesting) and for machine-to-machine ingestion (IoT feeds, scrapers, schedulers). It behaves identically in-process against the database and over HTTP against a remote instance. See the Python API page.
What is a versioned (bitemporal) time series?
A bitemporal series records two timelines: when a value applies, and when it became known. Every insert is kept, so you can query any series as it was known at any past date — for example, replay yesterday's price forecast exactly as it stood before this morning's revision. Computed series follow the same law: formulas evaluate against the versions known at the requested date.
Is The Timeseries Refinery really open source? Under what license?
Yes — the components are licensed under LGPL-3.0, and developed in the open in the Timeseries Refinery group on Heptapod. The free edition is published on PyPI; the pro components (dashboarding, series monitoring, access rights, MCP server) are open source as well, delivered with a subscription.
Can I self-host it, or is it SaaS only?
Both. Pythonian operates SaaS instances on EU-based cloud services with daily backups, or you deploy on your own infrastructure — MS Azure, AWS, Google Cloud Platform or plain Linux servers.
How long does deployment take?
A SaaS instance is operational in minutes. Self-hosted, the platform is Docker-based and operational in hours. Either way, no data engineering team is needed before analysts can start working with series.
Can AI assistants work with the data?
Yes. The platform includes an MCP server: any MCP-compatible AI assistant can search the catalog, query series, and build and evaluate formulas in natural language. The results follow the same rules as everything else — versioned and traceable.
Does it integrate with Excel?
Yes. The Excel client uploads and downloads series directly from the shared catalog, so teams keep their Excel workflows while the data itself stays versioned, named and traceable in the Refinery.
What data sources can it connect to?
An open source data catalog covers ENTSOE, ENTSOG, RTE, ECMWF, ERA5, Yahoo Finance and more. ARK connectors integrate commercial market data providers without development. Custom sources are integrated as Python tasks through the built-in scheduler.