About
Timestral Studio is built by one person who needed a faster forecast.
I am Charlie Simmon. I build and operate small AI web products. Most forecasting tools ask a team to learn a Python stack, train a model, or buy a platform before they can answer one question: what is this series likely to do next?
Timestral Studio is my answer to that question. Upload a CSV, choose a horizon, and get a median forecast with an 80% range and all 21 quantiles. The same model is available as a metered REST API for teams that need it inside a product or a nightly job.
Why this site exists
Mistral AI released Timestral-209M as an open, zero-shot time-series model. It is small enough to run without a GPU and useful enough to forecast a real business series without training a custom model. That combination is rare. I built a studio around it so a person can test the model on their own data in minutes, not after a procurement cycle.
The product is deliberately narrow. It is a CSV studio and an API, not a BI platform, a data warehouse, or an AutoML suite. If you need connectors, scheduled retraining, or a private deployment, the API and the annual contract path are the honest starting points.
What Timestral Studio is — and what it is not
Timestral Studio is an independent product. It uses the open Timestral-209M model from Mistral AI, but it is not affiliated with or endorsed by Mistral AI. When you search for “Timestral”, you may be looking for the model family; this site is the hosted studio and API built on top of it.
The model card and the public demo are the best way to inspect the underlying model. The studio adds CSV validation, a browser experience, credit metering, stored result files, and a documented API contract.
How the product works
Your CSV
A date column and up to eight numeric series. No training set, no schema migration.
The open model
Timestral-209M produces a median and 21 quantiles zero-shot. One credit per call.
Your output
The studio charts the band; the API returns JSON and a downloadable 21-quantile CSV.
Forecast calls are metered. A failed call is refunded automatically. The API contract, error codes and credit rules are documented in the API reference. The current pricing catalog is on the pricing page.
Data handling in one paragraph
The CSV you submit is sent to the model host to produce the forecast. Result files are stored in Cloudflare R2 so the link you receive does not expire with the model host. We do not use your uploads to train a model. Do not upload personal data or regulated data you are not allowed to send to a third-party inference host.
Contact
For support, a volume contract, or a correction to anything on this site, email contact@timestral.org. I read it.
Last updated: 2026-10-18