Data Analytics Suite · Code is the source of truth

Language models are very good at writing code. So let them.

Scripts, queries and dashboards are all code, kept as plain text files in the project. A brief that would take a day of clicking becomes a few hundred lines the agent writes, runs, fixes and runs again in minutes.

Join the waitlist

SQL over the simulation summary: simulated and observed oil rate by well — A query over the simulation summary: simulated and observed oil rate by well.

The agent works fastest and most reliably in code. That's why every computation, every table and every chart in Barrel Brain is a file the agent wrote.

Code is also what makes the work checkable. A chart in a spreadsheet hides how it was made. A script, a query or a dashboard shows every step. It can be reviewed, rerun on new data and reused on the next field.

Scripts · Compute

Short code that you can read

When the agent computes something, it writes a short C# script that reads from the workspace and calls the engineering libraries. You can open it, read it and run it again. The Scripting API Explorer lists the whole surface, so nobody guesses what's available.

Domain Decks, grids, summaries, wells, seismic, studies
Engineering The 15 calculation domains
Deliverables Excel, Word, PowerPoint
Application Files, processes, terminal

engineering_decline_rates.scriptC#

using Team76.Engineering.DCA;

public class S : IAppScript
{
    public ScriptResult Run(IScriptContext ctx)
    {
        // Turn a 15%/yr nominal decline into the
        // rates a forecast uses.
        double nominal = 0.15;

        return ctx.Result()
            .Set("effectiveAnnual",
                 DeclineRateConversions.NominalToEffective(nominal))
            .Set("nominalMonthly",
                 DeclineRateConversions.AnnualToMonthly(nominal));
    }
}

Queries · Analyse

SQL over every file in the project

A query treats the workspace as a database. Simulator output, well logs, seismic headers, CSV, Parquet, Excel and SQLite are all tables, and one statement can join them. It's how the agent pulls a hundred cases into one table, and how you check the table it pulled.

eclipse_summary Rates and totals by step
eclipse_grid Cell properties from EGRID and INIT
eclipse_restart Pressures and saturations per report step
eclipse_deck Keywords, sections and includes
las · segy Log curves and seismic traces
xlsx · sqlite · csv Everything else in the folder

oil-by-well.querySQL

-- Oil rate per producer, one row per well and date.
UNPIVOT (
    SELECT date, COLUMNS('^WOPR:')
    FROM eclipse_summary('KESTREL.SMSPEC',
                         vectors := ['WOPR:*'])
)
ON COLUMNS('^WOPR:')
INTO NAME vector VALUE oil_rate
ORDER BY vector, date;

-- Join simulator output with a spreadsheet.
SELECT w.well, w.field, sum(p.oil) AS oil
FROM 'production.csv' p
JOIN 'wells.parquet' w USING (well)
GROUP BY ALL ORDER BY oil DESC;

A query over the Excel workbook: cumulative oil and latest water cut by producer

Dashboards · Communicate

Results that stay current when the data changes

A dashboard is a page of sections, sliders, pickers, tables and charts, fed by queries, scripts and workspace files. When a file it read changes, for example after a new simulation run, it renders again. What you show your team is never a stale screenshot.

Data Queries, scripts and files as table sources
Pickers Well, vector and file pickers bound to the data
Table · Chart Bound to the same data
Office Word reports, PowerPoint decks, Excel workbooks

decline.dashboardmarkup + C#

<Section Title="Decline">
  <Slider Label="Initial rate" @bind-Value="Qi"
          Min="100" Max="2000" Unit="bbl/d" />
  <Slider Label="Decline" @bind-Value="D"
          Min="0.05" Max="0.5" Unit="1/yr" />
</Section>

<Section Title="Forecast">
  <Table Rows="Rows" Columns="Year, Rate:bbl/d" />
</Section>

@code {
  double Qi = 1000, D = 0.2;
  record Row(int Year, double Rate);

  // Exponential decline: q(t) = qi · exp(−D t)
  IReadOnlyList<Row> Rows => Enumerable.Range(0, 6)
    .Select(t => new Row(t, Qi * Math.Exp(-D * t)))
    .ToList();
}

Keep the tools you have

Python, IPM and your simulators, beside the C#

Python

If Python is installed, the agent writes and runs Python in the built-in terminal, so your team's scripts and packages keep working.

Petroleum Experts IPM

Scripts can drive PROSPER, GAP, MBAL, PVTp and REVEAL through OpenServer.

Any command-line simulator

Eclipse or OPM Flow run from a launch profile. The agent queues the runs, watches them and reads the results.

Fair question

What people ask about the code

“Our engineers don't write code.”

They don't have to. The agent writes the code, and engineers read results in viewers.

The code is there so the work can be checked and repeated, not so engineers have to type it. Most people will never open a script. The one who wants to know how a number was made can open it, read it and rerun it.

More of Barrel Brain

Agentic Harness → An agent harness for petroleum engineering.Office Productivity Suite → Excel, Word and PowerPoint from the tables you checked.Reservoir Simulation → Decks, grids, studies and results.