Build with agental.
A Linux machine of its own for every agent, over HTTP. Start with four calls, or go straight to a pillar.
Quickstart
- 1
Get an API key
Make one in the console under Keys, with the preset ci, and put it in AGENTAL_API_KEY.
- 2
Create a sandbox
A whole Linux machine. Keep its id.
- 3
Run a command
Anything a shell can run: Docker, a test suite, a browser.
- 4
Stop it
The disk is kept, at 5% of the running price. Start picks up where it left off.
curl -X POST https://agental.io/v1/sandboxes \
-H "Authorization: Bearer $AGENTAL_API_KEY" \
-H "Idempotency-Key: task-42" \
-H "Content-Type: application/json" \
-d '{"name":"agent-1","size":"small"}'How the API works
Every call starts here.
https://agental.io/v1Plain HTTP and JSON. Send bodies with Content-Type: application/json.
Every error has the same shape: {"error":{"code":"...","message":"..."}}. Branch on code.
A command that exits non-zero is still HTTP 200. Read exitCode in the body.
Creates and forks carry an Idempotency-Key, so retrying them is always safe.
| Size | vCPU | Memory | Disk | Per hour |
|---|---|---|---|---|
| tiny | 1 | 2 GiB | 20 GiB | $0.0135 |
| small | 2 | 4 GiB | 40 GiB | $0.027 |
| medium | 4 | 8 GiB | 80 GiB | $0.054 |
| large | 8 | 16 GiB | 160 GiB | $0.108 |
Leave size out to get small. Time is billed by the second: running at the full price, and stopped at 5% of it until the sandbox is deleted. The free plan makes tiny and small sandboxes, the $100 plan makes medium ones too, the $500 plan makes every size, the $2,000 plan makes every size. tiny has one vCPU, so one busy process holds up everything else in it: it suits work that mostly waits, such as an agent calling a model, a bot or a small service. Builds, Docker images and a browser want small or larger.
By pillar
Execution
A shell, a browser and a screen, over HTTP.
Persistence
Stop it, and the whole disk is kept.
Security
Keys, allowlists, and no runaway machines.
Tools
API reference →
Every endpoint, its fields and its errors.
GET /v1/sandboxesTypeScript →
The SDK for Node 20.16 or later.
npm install agental-sdkPython →
Sync and async clients, Python 3.10 or later.
pip install agental-sdkAgent skills →
Teach Claude Code, Codex and the others to use sandboxes.
npx skills add https://agental.io -gReading this with an agent?
Every page is Markdown too. Hand it llms.txt and it has all of them.
agental.io/llms.txt