Access by request

An AI CFD engineer for OpenFOAM.

In your terminal. Point cfdpilot at any OpenFOAM case. It reads your real setup and logs, explains what's going on, and acts on it: fixes as a diff you approve, then runs your solver. No upload, no browser.

Request access → Works over SSH · Python 3.8+
turbulent_pipe — cfdpilot
$ cd ~/simulations/turbulent_pipe
$ cfdpilot

CFDpilot Agent v0.1.20
──────────────────────────────────────────
Case: pimpleFoam · kOmegaSST · blockMesh
──────────────────────────────────────────

> why did my simulation diverge?
parse_log...

Co_max = 1.09 at t=0.043 s — Courant limit violated
k residuals: 2.3 → unstable after iter 17

Root cause: deltaT too large for mesh size.
Fix: system/controlDict

show_diff...
- deltaT 0.001;
+ deltaT 0.0001;

Apply changes? [Y/n] y

Patching system/controlDict · backup → .bak
Launching pimpleFoam...

Time=0.01 Co=0.09 k=1.2e-3 p=4.1e-4
Time=0.02 Co=0.09 k=8.1e-4 p=2.3e-4
Converged at t=0.18 s
<60s
from cfdpilot to first fix proposal
18+
solvers: simpleFoam · pimpleFoam · interFoam · rhoPimpleFoam…
8
tools: read · parse_log · run_solver · show_diff · patch_dict…
0
uploads — works directly on your files

Hours lost.
On every case.

01
You know it crashed. Not why.
The log says FOAM FATAL, Floating point exception, or solution singularity. You search CFD-Online for an hour. You get generic answers that don't match your mesh, your deltaT, your setup. You're back to square one.
02
ChatGPT doesn't know your files.
You paste your fvSchemes and ask what's wrong. It gives you textbook advice. It doesn't know your mesh, your deltaT, your actual Courant number. It can't.
03
Senior engineers are expensive.
The person who could diagnose this in 5 minutes costs $150/h. You spend 3 hours debugging what a trained model could catch instantly from your own log.
CFDpilot reads your actual files — no upload, no copy-paste, no guessing.
Request access →

One command.
Full context.

No prompt engineering. No copy-pasting files. CFDpilot reads everything from disk — your actual case, your actual log.

step_01

Set up in 30 seconds

Once you have access, a single command installs the CLI, then cfdpilot login. Works on your laptop, your workstation, or your HPC cluster over SSH. No config files needed.

step_02

Run in your case directory

cd your-case && cfdpilot. It auto-detects your solver (simpleFoam, pimpleFoam…), turbulence model (kOmegaSST, kEpsilon…), whether blockMesh ran, and which log file to parse. No config.

step_03

Ask, diff, fix, rerun

The agent parses your log into structured residuals, Co, and divergence events — not just raw text. Ask what went wrong. Review a colored diff. Confirm the patch. Residuals stream live until convergence or auto-stop on divergence.

Request access →

Not a chatbot.
A terminal agent.

It reads your files directly. It runs commands. It patches with your confirmation. It knows OpenFOAM cold.

F.01

Reads your actual case

The agent reads your case from disk — sandboxed to your case directory. It opens 0/, constant/, system/, and your log file directly. No upload, no copy-paste, no ZIP. Your files never leave your machine.

fvSchemesfvSolution 0/k · 0/Ulog.simpleFoam
F.02

Runs OpenFOAM commands

The agent invokes blockMesh, checkMesh, setFields, decomposePar, and your solver directly. Residuals and Co stream live to your terminal. It auto-stops on divergence — no babysitting needed.

blockMeshcheckMesh pimpleFoamfoamRun
F.03

Diff before every change

Every patch shows a colored diff first — for example - deltaT 0.001;+ deltaT 0.0001;. You confirm with Y or skip with N. A .bak is always written first. After 3 failed patch attempts, the agent stops and explains why.

show_diffpatch_dict .bak backup3-attempt limit
F.04

OpenFOAM rules baked in

Hard constraints, not suggestions: LES requires Co < 0.2 always. k-ω SST: y+ < 1 for wall-resolved, 30–300 for wall functions. GAMG: nCellsInCoarsestLevel ≥ 10. PIMPLE: nOuterCorrectors ≥ 2 for strong coupling. No linearUpwind on div(phi,U) in LES.

LESkOmegaSST GAMGPIMPLE
Request access →

What engineers say.

Early users running real OpenFOAM cases. Anonymized at their request.

“It successfully identified the mistakes in my setup, and the case now runs. I really appreciated how clearly it explained its reasoning.”

— CFD Engineer, Industrial

“Amazing, almost spooky in how it interprets and advises.”

— Aerospace Engineer

“It got my simulation to converge, and it explained the CFD reasoning behind each fix.”

— CFD Engineer, R&D

Request access.

CFDpilot onboards new users by request. Tell us about your case and we'll get you running the agent on your own OpenFOAM case.

Request Access
For the first engineers on board
Install in 30 seconds. Works on your server over SSH. Patches your files only when you say yes. No upload.
Request access → Python 3.8+ · Works over SSH

Common questions.

Once you have access, run cfdpilot login once, then cd into your OpenFOAM case and run cfdpilot. The agent reads your files directly from disk — fvSchemes, fvSolution, boundary conditions, solver logs — and gives you a diagnosis grounded in your actual numbers. When it finds a fix, it shows a colored diff and waits for your confirmation before patching anything.
Yes — CFDpilot is an AI assistant for CFD that works as a copilot for your OpenFOAM cases: it reads the solver logs, explains the error, and patches the setup from your terminal. No browser, no upload.
Python 3.8 or later and an OpenFOAM case directory in your terminal. That is it. No ZIP, no browser, no upload. CFDpilot reads your files directly from the filesystem.
Yes. CFDpilot is a pure terminal tool — it was designed to run on remote servers over SSH. Install it once on your HPC node and run it from any session. No GUI, no browser, no port forwarding needed.
Every proposed change shows a colored diff first. You type Y to apply or N to skip — nothing is ever modified automatically. Before patching, CFDpilot creates a .bak backup of the original file. The agent is sandboxed to your case directory and cannot touch files outside it.
CFDpilot supports 18+ solvers: simpleFoam, pimpleFoam, icoFoam, interFoam, rhoPimpleFoam, rhoCentralFoam, buoyantSimpleFoam, and more. Both OpenFOAM Foundation (v9–v12) and ESI (v2106–v2412) are supported.
On every run, the agent checks: Co_max violation (deltaT vs. mesh size), residual divergence pattern, y+ compliance for wall treatment, GAMG coarsening level, relaxation factors, scheme consistency (e.g. no linearUpwind on div(phi,U) in LES), boundary condition mismatches, and solver/turbulence model compatibility.
Your case files stay on your machine. The agent reads them locally and sends only the relevant content to the AI model to generate a diagnosis. Files are never stored on our servers. See our privacy policy for full details.

OpenFOAM troubleshooting guides.

In-depth articles covering the most common OpenFOAM errors and configuration questions.

View all 26 guides →

One command.
A CFD engineer in your terminal.

No upload. No copy-paste into ChatGPT. Just cfdpilot in your case directory.

Request access → Python 3.8+ · Works over SSH