Running in the Cloud

When you edit a project in the cloud editor, Visivo Cloud runs it for you — it executes your models against your sources, builds the insight and input data, and publishes the result when you commit. This page covers how that run works, what software the run environment ships, and the isolation model around it, so you know what your project code can rely on.

The cloud run loop

A cloud run is the same visivo run you know locally, executed on Visivo's infrastructure while you edit:

1. You edit in the cloud editor — each change is saved to a working draft.
2. Visivo runs the draft — it assembles your project and builds the data: models query your sources, seeds run their commands, and the insight and input data is produced. This happens as you work, not just at the end.
3. You commit — the built version goes live at its URL with fresh data and thumbnails.

While you're editing, cheap schema and reference validation runs continuously too, so you get feedback without waiting on a warehouse.

Deploying from the CLI does not run in the cloud

visivo deploy uploads a project you have already built locally with visivo run — the run happened on your machine, and Cloud simply hosts the result. Cloud runs are for projects you edit in the cloud editor; the rest of this page applies to those.

The run environment

The runner is the execution environment for your code: a seed's command runs arbitrary shell/Python, and a model's SQL runs arbitrary queries. Like a CI runner, it carries common tooling — but a deliberately curated subset, not a full CI image. The image is spun up per run, and a fat image would slow every run's start, so it stays slim.

Software	Version
Python	3.13
Node.js + npm	24.x
Visivo	matches the version serving your dashboards
Warehouse drivers	Snowflake, BigQuery, DuckDB, Redshift, ClickHouse, Postgres, MySQL (bundled with Visivo)
Shell tooling	git, curl, wget, ca-certificates, build-essential

Curated, not comprehensive

A hosted CI runner like GitHub's ubuntu-latest ships ~30 GB of languages and tools. The Visivo runner installs only what a project run needs, so if your seed shells out to something beyond the list above, install it in the seed command itself (pip install, npm i, apt-get install). Treat the versions above as the current baseline — they track the runner image and can move forward as it is updated.

Isolation and credentials

Because the runner executes your code, each run is sandboxed and holds no standing cloud credential — there is nothing ambient for code to reach for.

What your project code can rely on:

• Your account's source secrets. Store them as account secrets and reference them as ${ env.NAME } in your source config; Visivo injects them into the run's environment. This is also how a cloud source authenticates — a literal password in a source config is rejected in the cloud.
• The tooling above, and outbound access to the internet and your warehouses.

What a run cannot do:

• Reach any other run's output — runs are isolated from one another.
• Persist anything between runs, or run without bounded CPU, memory, and process limits.

A run naturally has access to its own account's credentials — it needs them to query your warehouses — so treat your seed and model code as trusted with your own data, exactly as you would a local visivo run. What's protected is the boundary between accounts, not between your code and your own sources.