---
title: "Quickstart: from nothing to a finished CFD case"
url: "https://tryreynolds.com/docs/quickstart"
description: "Make an account, connect a model, and ask Reynolds for a simulation. Covers both the browser app and the openreynolds command line, what the first study costs, and how to read the result it gives back."
---

**[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/docs/quickstart. It lives at https://tryreynolds.com/docs/quickstart.md, and
[tryreynolds.com/llms.txt](https://tryreynolds.com/llms.txt) indexes every one of them.

# Quickstart

Two ways to run a study. The browser needs nothing installed. The terminal needs Python and one command. Both end with the same thing: a case, a mesh, a solve, and the files that made the answer.

## In the browser

01

### Create an account

[app.tryreynolds.com](https://app.tryreynolds.com), email and a password. Confirm the address from the email, then sign in. The sign-in page also carries whichever one-click providers are enabled for the app; where **Continue with Google** is offered, it signs you in with no password to set or remember. Your workspace key is issued on first sign-in; there is nothing to copy or paste.

Sign up with your company email address and your account starts with $10 of credit, once. It covers the workspace and Reynolds’ model together, and it is a one-time credit rather than a monthly allowance. A personal address (gmail, outlook, icloud and the like) starts at zero; add credit from $5 before the first study runs.

02

### Connect a model

Pick Reynolds’ model, which is Claude metered to your account and needs no key of your own, or paste a key from your own provider. Either way the choice is checked before it is saved: the model has to answer, and it has to be able to see an image, because Reynolds looks at its own renders.

03

### Start a study and describe the flow

A study is a workspace, not a chat: it has an instance, a mesh, jobs and files, and it outlives anything you said to it. Name it, choose how far to take it, then write the first message in the study’s own composer while the workspace boots.

The same sheet asks how much the agent should ask you. **Full auto**, the default, leaves everything to it. **Ask before compute** puts each background job and mesh-desk build to you first. **Structured** has you approve a plan and then each stage. You can change it later from under the composer, so the first choice is not a commitment. [Modes](https://tryreynolds.com/docs/modes.md) explains each one.

![The Reynolds studies list: a left rail with Studies, Models and Account; a running session shown as a green band; and each study as a row with its figure, title, id, job and render counts, and controls to open, resume or replay it.](https://tryreynolds.com/assets/ui-app-studies-aYri89uV.jpg)

Every study you have run. A study keeps its instance, its mesh, its jobs and its files between sessions, so the list is of workspaces rather than of conversations.

## In the terminal

The command line runs the same session on your own machine, with the study files mirrored into the directory you started from.

```
pip install openreynolds        # or uvx, pipx, npm i -g openreynolds
openreynolds login              # email and password, the same account as the app
openreynolds config             # which model, and its key
openreynolds doctor             # check it can all be reached before spending anything
openreynolds                    # start a study
```

**Run `doctor` first.** It checks the settings file, the workspace service, the model API, capture, the toolbox, the terminal and the video encoder, and it writes nothing. Almost every confusing first session is one of those seven being wrong, and doctor says which in a sentence.

A terminal study starts in full auto. `openreynolds --mode partial` asks you before each background job and mesh-desk build, and `--mode structured` has you approve a plan and each stage. Inside the session, `/help` lists what you can type, and typing `/` then Tab completes a command.

## What to ask for

Describe the flow the way you would describe it to an engineer, in the units of the job. Geometry, the fluid, the speeds and temperatures that matter, and what you want to know at the end. You do not need to name a solver, a turbulence model or a mesh strategy. If you do name one, it will be used.

A first request that works well looks like this:

> Flow past a cylinder at Re 100 in water, 2D. I want the shedding frequency and a Strouhal number I can compare against the published value, and a picture of the wake.

That is enough for the agent to choose a domain, a mesh, a solver and a run length, and to know what it is being asked to report. What it is not is a specification for it to fill in.

## What the first study costs

Two meters run at once: the model, charged per token, and the compute, charged per second the instance is up. A short 2D case is cents. The two things that move the bill most are how hard the model is asked to think and how many cores the solve asks for, and both are settings rather than rewrites. [Models and cost](https://tryreynolds.com/docs/models.md) has the detail.

## Reading what comes back

Every study hands back the case directory, the solver logs, the mesh and the figures. When Reynolds reports a number it also reports where the number came from, and when it is not confident it says so: an unconverged residual, a mesh nobody checked, a Courant number higher than the scheme really wants. Those sentences are the useful part. A CFD answer is a claim about a case, and the case is right there to inspect.

## 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.

### Starting from a CAD file

The shortest real path is: create an account, connect a model, upload the STL, and
say what you want measured. The agent inspects the surface before it meshes it, so
an export with open edges, duplicate facets, mixed units or inverted normals is
something it discovers and works around rather than something you have to fix
first. The mesh tool picks the cheapest mesher that fits the shape: gmsh building the
body directly where the shape can be described with primitives and booleans, blockMesh
for a box-shaped domain, and snappyHexMesh or cfMesh where the geometry genuinely
arrives as a triangulated surface, which tolerates a surface that is not watertight.
Boundary layers are sized to the y+ the turbulence treatment needs.

You do not have to know any of those words to run this. They are here so that you
can check the answer afterwards.