# A computer for every agent task Run generated code, build previews, and collect results in an isolated Linux environment. Start fresh for the next task, or keep the workspace and come back. [Create your first sandbox](https://docs.watasu.io/sandboxes/overview/) [Try the first command](https://watasu.io/sandboxes#first-command) ## Open a shell with one command One SSH command opens a shell in a fresh sandbox. Authenticate with your Watasu API token the first time; your public key is registered for future connections. Building it into a product? Use the HTTP API or the SDKs to manage the same sandboxes from your application. [Explore the SSH interface](https://docs.watasu.io/sandboxes/ssh/) A real shell. No SDK required. ``` # A fresh sandbox from the base template ssh base@box.watasu.io # Run a command and receive its exit code ssh base@box.watasu.io 'python3 -V' # Reconnect to your existing sandbox ssh s-6231@box.watasu.io ``` With the CLI installed, `watasu ssh base` handles first-time token authentication for you. Create an environment from your application. ``` curl https://api.watasu.io/v1/sandboxes \ -H "Authorization: Bearer $WATASU_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "template": "base", "timeout": 3600, "lifecycle": { "on_timeout": "pause" } }' ``` Create a write-capable API key in the dashboard. The response includes the sandbox ID and data-plane connection details. ## Run the code Build the feature 01 / AGENTS ### A workspace for every task Let an agent inspect a repository, run generated code, create a report, or build a preview. Keep its work in a dedicated environment for the session. 02 / EVALUATIONS ### Start fresh, compare fairly Run tests and benchmarks from a pinned template. Execute jobs in parallel within your team limits, collect the results, and clean up after the run. 03 / YOUR PRODUCT ### Make execution a feature Build notebooks, user automations, coding tools, and plugin workflows. Give each job a real filesystem and process control behind an API. ## Run, pause Pick up again Run for the length of the task. Pause to release compute, or save a disk checkpoint to try another approach without starting over. 1. 01 ### Prepare `template` Capture the environment once. Choose a template and pin a version when reproducibility matters. 2. 02 ### Run `running` Give the task its own Linux environment, stream its processes, and collect what it produces. 3. 03 ### Branch `checkpoint` Save a disk checkpoint. Start a fresh sandbox from it to explore another path. 4. 04 ### Come back `paused` Pause to release compute and keep the machine. Resume before the retained disk expires. 5. 05 ### Finish `destroyed` Destroy the sandbox when the work is done. Retained checkpoints and templates have their own lifecycle. **Pause keeps more than the files.** A machine-state image can preserve memory and processes for resume. If that image is unavailable, Watasu restarts from the retained disk. Inactive disks expire after seven days. [Lifecycle details](https://docs.watasu.io/sandboxes/overview/) ## A workspace you can control Stream a process, inspect a file, expose a preview. Control the workspace through the same API you use to create it. 01 ### Real processes, live output Run commands, open interactive terminals, and stream stdout and stderr over live sockets. Get the exit code when the work is done. [Processes & terminals](https://docs.watasu.io/sandboxes/files-and-commands/) 02 ### Files your agent can work with Upload inputs, read and write files, watch for changes, and download results. Use plain HTTP or signed URLs to move the work. [Files & directories](https://docs.watasu.io/sandboxes/files-and-commands/) 03 ### The same environment, every run Build your packages and setup into a reusable template. Pin an immutable version when every run needs the same starting point. [Reusable templates](https://docs.watasu.io/sandboxes/templates/) 04 ### Share a working preview Expose an HTTP port and get an HTTPS preview URL. Share a working result while the sandbox is running. [Networking & previews](https://docs.watasu.io/sandboxes/networking/) 05 ### Save a checkpoint, branch the work Save a named disk checkpoint, then branch into a fresh sandbox from that state. Set an expiry to match the life of the task. [Disk checkpoints](https://docs.watasu.io/sandboxes/checkpoints/) 06 ### Decide what can connect Set internet and package-registry access, allow or deny specific destinations, and update network policy on a running sandbox. [Network controls](https://docs.watasu.io/sandboxes/networking/) ### Keep a workspace across tasks Named, team-scoped persistent volumes let later sandboxes mount the same workspace or cache. [Persistent volumes](https://docs.watasu.io/sandboxes/overview/#persistent-volumes) ## Pay for running time Compute is billed per second for the CPU and memory you request. Paused disks, checkpoints, and templates are billed separately as storage. vCPU €0.000020 / vCPU·s Memory €0.000004 / GiB·s [See all sandbox plans](https://watasu.io/pricing#sandboxes) EXPLORE A RUN 2 vCPU / 1 GiB Running time 90 seconds €0.003960runtime cost 30 seconds 1 hour Illustrative runtime cost at the published rates. Excludes plan fees and storage. Actual resource choices depend on your template and team plan. When compute stops: Retained disks **€0.10 / GiB·mo** Checkpoints & templates **€0.03 / GiB·mo** [Billing details](https://docs.watasu.io/sandboxes/billing/) ## Keep your agent’s workspace in the EU Watasu hosts sandboxes in EU data centers. Know where your agent’s workspace runs before you send the first task. [Security & data residency](https://watasu.io/security) ## Run the agent Host the product Use sandboxes for agent tasks and Watasu apps for your web service, workers, and APIs. Connect managed databases and storage to build the rest of the product. [Start building](https://dash.watasu.io/) [Explore the platform](https://watasu.io/) [FROM THE WATASU BLOG **Your terminal\ is the SDK** The story behind the SSH front door](https://watasu.io/blog/2026-07-07-ssh-into-sandboxes/)