earth-labs

your archive isa dataset now.

click anywhere on the crust · drill
lat 40.712800° / lon -74.005200° / depth 3214.7 m
agent · online
seismic record · live · channels 01—06sample rate 1 khz · gain 12 db
§ 01the problem
§ 01
the problem

decades of subsurface knowledge, locked in pdfs no one can query. we built the agent that reads them.

exploration companies sit on archives of well reports, core descriptions, and survey documentation — data that cost billions to acquire, stored in formats built for printing, not analysis. most of it is never queried again.

today's options: hire $250k+ ml engineers to build one-off parsers, commission a months-long digitization project, or let it stay dead. every drilling decision made without the archive is a more expensive decision.

§ 02the product
§ 02
the product

the archive agent — point it at your file server. get answers.

ingest · classify · extract · validate · query
in
any exploration pdf — well headers · core & cuttings descriptions · survey reports · completion reports. no integration project, no workflow changes.
agent
classifies every document, extracts each fact into OSDU-compatible schemas, and runs physics-plausibility checks on every value — a porosity of 85% in granite gets flagged for review, not silently stored.
out
a queryable store you talk to in plain english: “every well in this basin with sonic and density logs below 2,000 m” — answered in seconds.
→ request a pilotdeploys in your vpc · your data trains nothing
§ 03why now
§ 03
why now

the archive problem, by the numbers.

corpus coverage · equirectangular10 active basins · 25.0 PB
-180°-120°-60°0°60°120°180°-60°-30°0°30°60°diskosnorth sea · 22 PBtnonetherlands · 5,800eposeurope · 800 TBnibiusa · 182,193aus.gov4,000 km 2D seismiciodp/lilypacific · 2 PBiodp/lilyindian oceaniodp/lilyatlanticiodp/lilysouth atlanticgeoph.asia
0
public wellbores · diskos alone
one archive of many — iodp, namss, epos, boem, and every private file server
0s of PB
subsurface data worldwide
seg-y · las · dlis · scanned reports — most of it unqueryable today
$250k+
the hire you don't make
structured extraction at a fraction of one ml engineer's salary
0
workflow changes
point at the file server · runs in your vpc · answers in days, not quarters