Docs / Get started
How EsaSim works
The few ideas you need: projects and blueprints, the workers that run them, and where everything lives.
EsaSim is where you build a simulation, run it on workers you choose, and use what it finds. Five ideas cover almost everything; the rest you meet as you go further.
The five ideas
| Idea | What it is |
|---|---|
| Project | One piece of work: one robot, one dataset, one study. It holds blueprints, runs, results and a team, and it is saved as a GitHub repository of yours. See Projects. |
| Blueprint | Blocks wired together on a grid: a simulation, the steps that process its output, and views that show the results. Also called a flow. See Blueprints. |
| Block | One step. Worker blocks are Python and run on your workers; views are JavaScript and draw in your browser. EsaSim has many built in, and you can write your own. See Build your own blocks. |
| Worker | The EsaSim worker: a program you install on your laptop, a workstation or a GPU server. It runs the worker blocks of your runs there. See Set up a worker. |
| Run | One execution of a blueprint. You watch it live on the blueprint's View page, and keep what it made under Results. See Runs. |
What happens when you press Run
- EsaSim takes the blueprint's published version and splits it into tasks, one per worker block.
- A worker that fits each task (the right GPU, Python, permission to run code) picks it up.
- The worker reads its input data where it already is, runs the block and streams progress, logs and live values back.
- Views fill in as values arrive. When a block ends, the next one starts; the results stay on the worker, and small summaries come to EsaSim.
Where everything lives
| What | Where |
|---|---|
| Blueprints, block code, readme, pictures, history | Your GitHub repository |
| Datasets | Your workers, buckets and cloud drives |
| Results of runs | The worker that made them (small summaries on EsaSim) |
| Trained policies you publish | Releases of the project's repository |
| Accounts, teams, run history, the queue | EsaSim |
Workers connect out to EsaSim over HTTPS; EsaSim never connects in. Nothing on your network has to open, workers work behind home routers and company firewalls, and your data is never uploaded to run a blueprint.
Three ways to build
- With an AI assistant. Connect Claude, ChatGPT, Cursor or Claude Code and describe what you want to learn; it builds the blueprint, runs it and reads the results with you. Tutorial.
- In the editor. Add blocks from the menu, wire them and set them in the browser. Blueprints.
- In code. Write blocks in Python and JavaScript and blueprints as JSON, in your own editor and git. Tutorial.
All three make the same project: you can switch between them at any time.
As you go further
| Idea | What it is |
|---|---|
| Workspace | Who owns the projects and workers: your personal workspace, or an organization shared with a team, with roles. See Organizations and roles. |
| Dataset | A folder, or some of its files, that a worker shares with EsaSim. EsaSim keeps its index and versions; the files stay where they are. See Data and datasets. |
| Published version | A blueprint's numbered, unchanging version. Every run uses one, so every result can be traced to what made it. |
| Published policy | What a blueprint trained, sent to GitHub as a release when you are happy with it. See Publishing trained policies. |
| Endpoint | A published blueprint your own app, service or CI calls over HTTPS with a key. See API endpoints. |
| Compute unit | A slice of a shared multi-GPU server: some GPUs, memory, disk and cores. See Compute units. |
In development: cloud GPUs on prepaid credits, for teams without hardware and for runs larger than their workers; EsaSim's own models and an API for agents; single sign-on.