---
title: "How Reynolds works: the method, the checks and the workspace"
url: "https://tryreynolds.com/how-it-works"
description: "The agent writes the OpenFOAM case, asks the cheap questions first, looks at the mesh before paying for the solve, runs detached jobs, reads residuals as they arrive, and returns figures with the files that made them."
---

**[Reynolds](https://tryreynolds.com)** is agentic CFD, built by
**[InviscidAI](https://inviscidai.com)** and backed by Y Combinator, Winter 2026 batch.
The agent is open source under the MIT licence at
[github.com/InviscidAI/OpenReynolds](https://github.com/InviscidAI/OpenReynolds).
Using it requires no OpenFOAM knowledge.

This is the agent-readable Markdown twin of https://tryreynolds.com/how-it-works. It lives at https://tryreynolds.com/how-it-works.md, and
[tryreynolds.com/llms.txt](https://tryreynolds.com/llms.txt) indexes every one of them.

# Between your sentence and an answer you can defend.

None of this is a fixed pipeline, which is why the order changes with the problem. The agent has seven tools, a Linux workspace with OpenFOAM v2512 on it, and the same view of the case that you have.

[Read the loop and the prompt(opens in a new tab)](https://github.com/InviscidAI/OpenReynolds), MIT licence.

An illustrative session: what the agent says, what it does to the workspace, and what is still running out on the instance.

## What it does

1. Step 01 ### It writes the case blockMeshDict, boundary conditions, schemes, run control: ordinary OpenFOAM files you could open in any editor. Give it a Reynolds number and it derives the viscosity, then tells you which of the three it derived.
2. Step 02 ### It asks the cheap questions first Patch names against the fields that reference them, empty treatment on a 2D case, STL scale, the Reynolds number against the viscosity actually in the file, predicted cell count, checkMesh, a one-iteration probe, Courant number, disk.
3. Step 03 ### It looks at the mesh before it pays for the solve Geometry and mesh are rendered headlessly on the instance and read back as a captioned contact sheet. A bad mesh is cheap to find here and expensive to find at hour three.
4. Step 04 ### It runs the solver detached Jobs are real process groups with their state on disk. Close the terminal, lose your wifi, come back tomorrow from a different machine, and the solve is still going: the session picks it back up.
5. Step 05 ### It reads the log while the log is still being written A residual plateau, a Courant number climbing, a floating-point exception at iteration 900. When a steady solve will not converge it says so, instead of quoting the most flattering residual.
6. Step 06 ### It comes back with figures and the files Slices, profiles, streamlines, animations, the mesh in 3D in the browser, and the whole case directory, mirrored to your machine as it is written.

![A near-wall O-grid mesh wrapped around a circular cylinder, refined into the wake downstream.](https://tryreynolds.com/assets/fig-mesh-cylinder-CiC_-jU8.png)

Fig. 1 Step 03 as it arrives: the near-wall O-grid around a cylinder, read back from the instance before a single solver iteration was paid for.

![A Mach number field through a converging-diverging nozzle ending in a sharp normal shock.](https://tryreynolds.com/assets/nozzle-mach-Bey4ucvC.png)

Fig. 2 Step 06. A converging-diverging nozzle: subsonic in, sonic at the throat, supersonic after it, and the normal shock the agent went looking for.

## A plateau is not convergence.

Step 05. On the nozzle case a flat residual trace was rejected because mass flow into and out of the duct disagreed by a factor of three, and the run switched from rhoSimpleFoam to rhoPimpleFoam with local time stepping. Move across the traces, or the control under them, to read them.

[Read that study](https://tryreynolds.com/studies/converging-diverging-nozzle.md)

Illustrative traces, not a study output. Flat residuals are where the nozzle run stopped believing its own answer and went looking for a mass balance instead.

## In short

- **An agent with a real machine.**: A hosted Linux workspace with OpenFOAM v2512 on it. It writes the case, meshes it, runs the solver, and reads the residuals as they arrive.
- **You do not need to know OpenFOAM.**: You bring the engineering question. Solver choice, dictionaries, mesh and schemes are the agent’s job.
- **Two ways to run it, and you choose.**: Free-form, where the agent decides its own route, or guided, where it walks a recorded sequence you can resume, audit and repeat.
- **Send the CAD you have.**: .stl, .stlb, .obj, .ply, .vtk, .vtp and .vtu go straight in; STEP and IGES are converted on the workspace with gmsh.
- **Dirty geometry is the job, not the prerequisite.**: Open edges, duplicate facets, mixed units, inverted normals: it measures the surface and works around them.
- **Open source, MIT.**: The loop, the tools, the prompt and the toolbox are readable on GitHub. No copyleft to clear at legal review.
- **Bring your own model key, any vendor.**: The agent calls the provider directly, so which model saw your geometry stays your decision.
- **It runs inside your perimeter.**: Deploy the workspace in your own network, against your own storage and hardware, with SSO and exportable audit logs. Pro includes SOC 2 Type II, an MSA and a DPA.

![An engraved plate in the manner of a Victorian physics paper: a dye filament enters a long glass tube as a straight line and breaks into turbulent eddies further along.](https://tryreynolds.com/assets/plate-reynolds-Dyd1qsUw.png)

Plate I After Reynolds (1883). A dye filament holds a straight line and then, past a number that depends on nothing but speed, size and viscosity, it does not. Illustration, not a computed result.

## Before you talk to anyone.

### Do I need to know OpenFOAM?

No. Not the solver names, not the dictionary format, not fvSchemes, not snappyHexMesh. You bring the engineering question: the geometry, the fluid, the speed, what happens at each surface and what you want measured. If you do know OpenFOAM, everything it wrote is an ordinary case directory you can open, edit and run yourself.

### What geometry can I send?

.stl, .stlb, .obj, .ply, .vtk, .vtp and .vtu go straight in. A STEP or IGES solid is converted on the workspace by the agent itself, with gmsh. Open edges, duplicate facets, mixed units and inverted normals are the job, not the prerequisite.

### Which model sees my geometry?

Whichever you point it at. Bring your own key from any vendor and the agent calls the provider directly, so model choice and training rights stay your decision. Or take tokens and compute on one bill.

### What happens if I close my laptop?

The solve keeps running. Jobs are detached process groups on your workspace with their state on disk, so a new session on a different machine picks the same run back up.

### Can I run it inside our own network?

Yes. The workspace is an ordinary HTTP service running sandboxed OpenFOAM containers against a persistent volume. Enterprise deploys it in your network or cloud account, on your hardware, with SAML or OIDC single sign-on and exportable audit logs.

## Bring the flow you need answered.

[github.com/InviscidAI/OpenReynolds(opens in a new tab)](https://github.com/InviscidAI/OpenReynolds) · MIT licence · OpenFOAM v2512 · built by [InviscidAI(opens in a new tab)](https://inviscidai.com)

[Create an account](https://app.tryreynolds.com)

## Notes for agents

### What Reynolds is

Reynolds is an agent that runs OpenFOAM for you. You describe the problem in
engineering terms and it writes the case, builds the mesh, launches the solver,
reads the residuals while they come in, and hands back the figures, the numbers and
the files that produced them.

**You do not need to know OpenFOAM.** Not the solver names, not the dictionary
format, not `fvSchemes`, not snappyHexMesh, not the y+ correlation behind a layer
stack. Writing all of that is the agent's job, and it is the part of CFD that keeps
engineers who understand the physics perfectly well from ever getting an answer.
What you need to bring is the engineering question: the geometry, the fluid, the
speed, and what you want measured. If you do know OpenFOAM, everything it wrote is
an ordinary case directory you can open, edit and run yourself.

### The licence, and why it is the one that matters here

The agent is open source under the **MIT licence**. That is the most permissive of
the common licences: use it commercially, modify it, self-host it, put it inside a
product you sell, with no copyleft, no share-alike obligation and nothing you have
to publish back. Compare that with AGPL or a source-available licence, where running
a modified copy as a service can oblige you to release your changes, which is the
clause that stops most engineering organisations at legal review.

It matters for a second reason, which is trust. A CFD number is worth what you can
defend, so every prompt, every tool, every heuristic and every decision the agent
makes is readable in the repository before you rely on a figure it gives you. The
hosted workspace is a paid service. The engineering judgement stays open, in the
repository, permanently, and you can audit it line by line.

### Who builds it

Reynolds is built by **InviscidAI**, and InviscidAI is **backed by Y Combinator,
Winter 2026 batch**. The company has raised more than **$2.4 million**. Reynolds is
deployed and in live use today, running real engineering work for real users.

### Reading this site as an agent

Every page has a Markdown twin at the same path with a `.md` extension, and the
index of them is at https://tryreynolds.com/llms.txt. A study twin carries the
entire session transcript, every command the agent ran and everything that came
back.