Vyom 1.0 · open for testing

Every field in India,
read from orbit. Even under cloud.

Vyom is an Earth-observation platform for agriculture. Outline a field with one tap, follow its season pass by pass, see it through monsoon cloud, and take all of it into your own research through an API.

Free while in testing · 25 analyses a day · sign in with an email code or a passkey

Held-out example
Vyom's reconstruction of the same pass Sentinel-2 as delivered, under cloud VyomSentinel-2, cloudy
NDVI error under the cloud, last clear pass–
NDVI error under the cloud, Vyom–

Drag to compare. Error against a clear pass a few days later.

0.098NDVI error under 40%+ cloud, 47 held-out scenes (last clear pass: 0.127)
0.837field-boundary F1 on 288 held-out 0.5 m images
896Vyom features per field, from a 47M-parameter foundation model
3sensors: Sentinel-2 optical, Sentinel-1 radar, 0.5 m basemap
Proof, not a promise

A cloudy pass, what Vyom rebuilt, and what was really there.

Every image here is a real output on a scene Vyom never trained on, shown beside the clear pass taken a few days later. Nothing is retouched.

Sentinel-2 as delivered
Sentinel-2 as delivered
Vyom's reconstruction
Vyom's reconstructionfrom the last clear pass + radar
The clear pass a few days later
What was there
This scene: NDVI error inside the cloud
––

Last clear pass carried forward, then Vyom. Measured on – pixels under the cloud mask.

All 47 held-out scenes with at least 40% cloud
0.1270.098

R² 0.45 → 0.74. Vyom was closer on 36 of 47 scenes; on the other 11, carrying the last clear pass forward was closer. Not yet tested in July to September.

Read the full results →
The platform

From the whole planet down to one bund.

Vyom Earth is a globe you fly into. Search a village, tap a field, and every analysis runs on that outline. The console keeps your fields, history, keys and usage in one place.

Vyom Earth

A 3D globe with satellite imagery, place names and roads. Search by village and district, a coordinate or a map link, and fly straight to the field.

Indiaat 0.5 m, Sentinel at 10 m

See through cloud

A cloud-free image and NDVI for the date you ask, rebuilt from the last clear pass and the radar change since then.

0.098NDVI error (0.127 before)

One-tap field outline

Tap inside a field and a boundary model reads 0.5 m imagery and returns its edge along the bund. You can edit every corner.

0.837boundary F1, 288 test images
New

Season time series

Every Sentinel-2 pass over the field for up to 400 days, with five indices, the spread inside the field, radar, and each season's green-up, peak and senescence.

CSVfor your own analysis

Six spectral states

Each 10 m pixel is placed in one of six rule-defined states, from low canopy water to a canopy drop against its own recent history. A reason to look, not a diagnosis.

0.584macro-F1 nowcast under cloud; exact rules when clear
New

Research assistant

Ask how a field behaved and why. It reads that field's stored results, explains the physics behind each index, and keeps to measured numbers.

Groundedin your field's own record
0.5 m imagery of farmland in Kolhapur, Maharashtra Field boundaries predicted by Vyom's boundary model Kolhapur, Maharashtra · held-out test image, boundary model output
Picked at the 25th percentile of per-image F1 (0.781), not a best case
Fields, not pixels

Smallholder plots are tiny. Vyom starts from the bund.

Most Indian fields are a few Sentinel-2 pixels wide. Analysing the right outline is the difference between a field's signal and its neighbour's.

  • Tap, check, save. The outline comes from 0.5 m imagery and stays editable corner by corner.
  • Up to 50 fields per account, each with its own history of analyses and season curves.
  • Turn the boundaries off on the image to see the imagery the model read.
Deeper than advice

Study how a field behaves through a whole season.

A season time series reads every pass over your field and keeps the clear ones. Each value is computed from that pass's own reflectance over the field's 10 m pixels. No model guesses, no crop named.

IndexFormula (Sentinel-2 bands)What it responds to
NDVI(B8 − B4) / (B8 + B4)Green canopy cover. Rises at green-up, peaks near flowering, falls at senescence; saturates in dense canopy.
NDRE(B8A − B5) / (B8A + B5)Chlorophyll in the canopy via the red edge. Keeps responding where NDVI saturates. Related to, but not a measure of, nitrogen.
NDMI(B8 − B11) / (B8 + B11)Canopy water, from short-wave infrared absorption. Separates a water problem from a canopy problem.
EVI2.5 (B8 − B4) / (B8 + 6 B4 − 7.5 B2 + 1)Greenness corrected for soil and haze; saturates later than NDVI.
MNDWI(B3 − B11) / (B3 + B11)Open water, such as flooded paddy or waterlogging (above 0).
VV, VHSentinel-1 backscatter, dBCanopy structure and surface moisture, through cloud and rain. Fills the monsoon weeks optical cannot see.
Pass by passMean, 10th and 90th percentile inside the field, so you see variability, not one number.
Seasons foundGreen-up, peak and senescence where the NDVI curve crosses halfway between trough and peak.
Gaps reportedThe longest stretch without a clear look, so you know where the curve is thin.
Yours to analyseDownload as CSV, or call it from code and get a job you can poll.
Who it's for

Built for the people who need to understand fields, not just be told about them.

01 · RESEARCH

Researchers

  • Season curves and radar per plot, as CSV or JSON
  • 896 Vyom features per field for your own models
  • Measured errors and test conditions you can cite
02 · LEARNING

Students

  • Learn remote sensing on real Indian fields
  • Compare rainfed and irrigated plots, kharif and rabi
  • An assistant that explains the physics behind each index
03 · FIELD TEAMS

FPOs

  • Watch every member plot from one map
  • See which plots changed, even after a cloudy week
  • A dated record per plot, before anyone drives out
04 · INSTITUTIONS

Universities, KVKs and departments

  • Monitor trial plots and whole blocks
  • Teach with a live platform and API keys for projects
  • Usage tracked per key, for every project
Sphitih and Vyom

Advice lives in the Sphitih app. Vyom is where you go deeper.

Sphitih app

For farmers: what to do this week.

  • Crop plans and a week-by-week calendar
  • Pest advisory from Sphitih's advisory agents
  • Only doses the Government of India registers for the crop
  • Equipment rental
sphitih.in →

Vyom

For understanding: how fields behave, and why.

  • Cloud-free imagery and NDVI on any date
  • Season time series with five indices and radar
  • Pixel-level spectral states and foundation-model features
  • A research assistant grounded in your field's record
  • API keys, per-key usage, history and exports
Start with Vyom →
API

A polygon in, a field's record out.

Send a GeoJSON polygon and a date. Get cloud-free NDVI, spectral states, map layers, a GeoTIFF and Vyom's 896 features. Season analyses run as jobs you poll.

25 / dayanalyses per account, shared by its keys
500 haper request
Per keyrequests, analyses, errors and latency in your console
OpenAPIreference at /docs
Get your API key → API reference
curl -X POST https://vyom.sphitih.in/api/v1/analyze-parcel \
  -H "X-API-Key: $VYOM_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "geometry": {"type": "Polygon", "coordinates": [[[76.8161,25.1838],
                 [76.8170,25.1838],[76.8170,25.1846],[76.8161,25.1846],[76.8161,25.1838]]]},
    "date": "2025-10-15",
    "tasks": ["cloud_free_ndvi", "spectral_state_6class", "embeddings"]
  }'
Start with one field

Mark it once. Watch it all season.