How to get 10 m embeddings with GeoTessera
How it is run today: Published data layer (not weights we run)
GeoTessera is how you get the published TESSERA tiles: annual ~10 m, 128-D embeddings. You query an area and a year (and a dataset version). You do not send new imagery into a hosted TESSERA model.
That fetch is CPU, storage, and bandwidth. No GPU is required for the published layer. Running the foundation model on new imagery is a different, self-host problem; GPU class for that path is unknown.
We can stand up a pass-through over the published embeddings. We do not already host that proxy, and we are not running TESSERA weights. Confirm the LICENSE in the upstream repo before a commercial wrap.
Upstream
Also in the catalog : TESSERA / GeoTessera.
Reach out
If this is the job you need, say so. We will not claim we already host it. We will say whether we pass through, stand it up, or cannot do it honestly.