THE BIGGER SYSTEM

Subsurface AI: a path to subsurface superintelligence

Today: reservoir simulation with CLARISSA. Our vision: subsurface AI that improves through verified engineering work.

Agentic reservoir simulation: from question to tested results.

CLARISSA connects an engineering request to model authoring, validation, simulator execution and analysis. Explore the AI reservoir simulation technology behind the workflow, including OPM Flow execution and the RIGOR benchmark.

01 · ASKTell CLARISSA what you want to model.
02 · BUILDDraft and validate the reservoir model.
03 · RUNRun the simulator and inspect the evidence.
04 · EVALUATEMeasure engineering performance with RIGOR.

Turn verified experience into better AI.

Verified workTraining examplesOptimizationStronger specialists
Development path: model specialization and full weight fine-tuning are future capabilities the data and verification infrastructure is being built to support. They are not presented as demonstrated production capabilities today.

Connect the geology to the engineering.

GEOLOGYStart from supplied geological descriptions, data or an existing geomodel.
RESERVOIR MODELTranslate interpretation into a structured simulation case.
SIMULATION INSIGHTConnect model behavior back to the engineering decision.

Deeper geology integration is part of the future vision. Today, CLARISSA starts from supplied geological descriptions or an existing geomodel.

Three tools. One connected system.

CLARISSA

The engineering agent

Turns an engineering request into a reservoir model, runs simulations, and helps analyze the results.

PetroScript

The model language

Turns structured engineering instructions into simulator inputs through consistent, deterministic compilation.

RIGOR

The benchmark

Measures how well an agent authors, repairs, runs, and analyzes reservoir simulations against verified tasks.

The model can change. The engineering foundation keeps growing.

Frontier model

Start with general-purpose model capability.

CLARISSA harness

Author, validate, run and analyze domain tasks.

Simulator evidence

Capture successful and failed trajectories with hard engineering feedback.

RIGOR

Measure performance and provide explicit reward signals.

Optimize

Improve prompts and trajectories today; support future model specialization.

Subsurface specialist

Use stronger specialists to help create better examples for the next generation.