Reservoir
Reasoning
Technologies
The instruments.
Clarissa, At Work
Footage in production
CLARISSA
An end-to-end reservoir simulation agent.
Clarissa writes simulator decks from scratch. It extracts the assumptions from your case, validates the deck, runs the simulation, and reads the results back as engineering insight — not raw output. Under the hood it speaks the simulators’ own languages and checks its own work at every step.
The loop, end to end
The point is not automation for its own sake. It’s that the craft of a senior reservoir engineer — which assumptions matter, what a healthy run looks like, when a result should not be trusted — becomes something your whole team can call on, on demand.
Measuring the Machines
Footage in production
RIGOR
A first-of-its-kind, open-source benchmark for reservoir-engineering agents.
As agents arrive in reservoir engineering, the field needs a common, trustworthy way to measure them. RIGOR is that yardstick: a public benchmark that puts agent harnesses through hand-authored OPM Flow tasks — writing simulator decks from scratch, modifying and debugging them, and analyzing results.
Scored on
- Deck validity — is the generated deck well-formed and runnable?
- Semantic requirements — does it capture what the case actually asked for?
- Successful runs — does the simulation complete?
- Output match — do the results line up with the gold reference?
- Analysis correctness — is the engineering read-back right?
It’s built for integrity. Every task is hand-authored and expert-reviewed; gold references stay verifier-only; and a “naked” baseline isolates the real lift from guidance, skills, and tools — so the numbers mean something. Open-sourcing soon, so the whole field can measure progress on the same footing.
The Language of Decks
Footage in production
PETROSCRIPT
The deck-authoring engine beneath Clarissa.
Simulator decks were never meant to be written by hand — thousands of lines of position-sensitive keywords where one silent mistake costs a study. PetroScript replaces that with a fluent, strongly-typed way to describe a reservoir model and compile it to a valid simulator deck.
- Fluent & typed — a model is composed step by step, and ill-formed models are rejected before a deck is ever emitted.
- Validated in depth — a registry of engineering checks runs at every layer, from units to schedule consistency.
- Reproducible — the same model compiles to the same deck, byte for byte. Studies stop depending on who typed them.
- Round-trip — results come back through the same library, so analysis sits next to authorship.
Clarissa writes on PetroScript the way an engineer writes on a keyboard: every deck the agent produces is one a machine has already checked — and one a human can still read.