API

Ask for the slice, not the loaf

Every request is a subset: variables, levels, times, an area and optionally an ensemble member.

$ isobar get --dataset ecmwf-ens --var 2t --member all \

$ --bbox 5,45,20,56 --time 2026-08-04T00/2026-08-11T00

resolving … 51 members × 169 steps × 1 variable

subset: 4.1 GB (full cycle would be 6.4 TB)

written to 2t_ens_20260804.zarr in 38s

Endpoints

GET/v1/datasetsList datasets with extents and variables

GET/v1/datasets/{id}/cyclesAvailable cycles and their status

POST/v1/subsetCreate a subset request, returns a job

GET/v1/jobs/{id}Job status and output URLs

GET/v1/zarr/{dataset}/{path}Direct Zarr chunk access

GET/v1/grib/{dataset}/{cycle}/{field}Original GRIB2 message

POST/v1/subscriptionsWebhook on new cycle availability

GET/v1/lagCurrent ingest lag per model, unauthenticated

Examples

python

import xarray as xr

ds = xr.open_zarr(
    &#x27;https://api.de-dev-01.tabserv.link/v1/zarr/era5/single-levels',
    storage_options={&#x27;headers': {'Authorization': f'Bearer {KEY}'}},
)

# 40 years of hourly 2m temperature at one point: ~200 ms
series = ds[&#x27;t2m'].sel(latitude=52.5, longitude=13.4,
                       method=&#x27;nearest')

json

{
  "dataset": "ecmwf-ens",
  "cycle": "2026-08-04T00",
  "variables": ["2t", "10u", "10v", "tp"],
  "members": "all",
  "steps": "0/168/3",
  "area": [56, 5, 45, 20],
  "format": "zarr"
}

bash

isobar auth login --key $ISOBAR_KEY

isobar get --dataset era5 --var t2m,tp \
  --time 1990-01-01/2025-12-31 --freq hourly \
  --bbox -10,35,30,60 --format netcdf --out ./era5_eu/

# 112 TB · resumable · checksums verified per chunk

Large jobs are expected. Requests above a terabyte are queued, chunked and delivered with a manifest; they resume cleanly and verify per chunk. There is no penalty for asking for a lot — the archive exists to be read.