Introducing Simular Platform: Use the Sai API to Give Your Agents a Fleet of Autonomous Computers

What is the Sai API?

The Sai API is the autonomous computer API of Simular Platform. It lets your code or your coding agent hand a task, in plain language, to Sai, a computer-use agent that does the work on a real computer and reports back the result.

Three parts work together behind one API key:

You reach it over REST, through the Python and TypeScript SDKs, or from Claude Code, Codex and Cursor over MCP. Sai is model-agnostic: you choose the model it plans with, and the model and computer are included in your plan, so there are no model keys to bring.

How it differs from a computer-use model API. A model API returns actions, and your application has to run them on a machine you host. The Sai API does the task itself: the computer, the agent loop, approvals and retries are included, and you read the result.

How it differs from a sandbox. A sandbox gives you a machine to script. The Sai API gives you a computer with an agent already operating it, so you send the goal instead of every click.

Your first task

Sign in at platform.simular.ai and create a key at platform.simular.ai/api-keys. The portal is where you sign in, manage keys and read the docs; requests go to the API host, https://api.simular.ai.

export SAI_API_KEY="sapi_..."

curl -s https://api.simular.ai/v1/agents/message \
  -H "Authorization: Bearer $SAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"message": "Open Notepad, type hello, save it to the desktop as hello.txt", "wait": false}'

{ "sessionId": "cs_01HZY", "machineId": "free-3fA9kQ2xYz", "queued": false }

Then poll for progress. Each response returns a cursor; pass it back as since and repeat while status is running:

curl -s "https://api.simular.ai/v1/agents/events?sessionId=cs_01HZY&wait=30" \
  -H "Authorization: Bearer $SAI_API_KEY"

{
  "events": [{ "type": "finish", "finishReason": "stop" }],
  "cursor": "1790000012345:9",
  "status": "idle",
  "text": "Done. hello.txt is on the desktop with the text \"hello\"."
}

On a free account, the first task of the day can take about two minutes while the computer starts. Keep polling; the session is not stuck.

What you can do

• Hand off tasks with no API. Log in to a vendor portal and list last month’s invoices, enter a CSV into a form with no import, or export users from an admin console.
• Test what a real user sees. Sign up on your staging site, click the verification email, and report where the flow breaks. Reproduce a bug report step by step.
• Get typed results. Add options with an output schema, and the turn returns JSON your code can read directly, with a record of each step.
• Stay in control. Sign-ins, credentials and other sensitive steps come to you as approvals. Answer them over the API, or open the approval link in a browser.
• Pay less on repeat runs. Run a task once, then save it as code (research preview). Later runs replay the code and ask the model only where the screen matters, which cuts tokens by up to 90% on repetitive tasks. When the UI changes, the agent takes over from that step.

How a task flows

1. POST /v1/agents/message with the task. With wait: false, the API answers at once with a sessionId.
2. Sai works on the computer. GET /v1/agents/events returns progress, text and approval requests.
3. When Sai needs permission, status becomes needs_approval. Answer with POST /v1/agents/approve, or send the person the approval link for sign-ins and credentials.
4. When status is idle, text holds the result. Follow-up messages continue the same conversation on the same computer.

Prefer a live stream? Send wait: true (the default) and the same events arrive over Server-Sent Events.

Three ways to call it

HTTP, from any language. Every request carries Authorization: Bearer sapi_... and goes to https://api.simular.ai. Check a key with:

curl -s https://api.simular.ai/v1/agents/auth \
  -H "Authorization: Bearer $SAI_API_KEY"

SDKs for Python and TypeScript. They handle polling and map errors to exceptions. Both are at version 0.1.0 and may change before 1.0.

pip install sai-sdk               # Python 3.10+
npm install @simular-ai/sai-sdk   # Node.js 18.17+

from sai import Sai

result = Sai().sessions.create().run("Open Notepad, type hello, save it to the desktop as hello.txt")
print(result.text)

Your coding agent. Claude Code, Codex and Cursor call Sai through the @simular-ai/sai-mcp server (Node.js 22.12 or newer). One command adds it to Claude Code:

claude mcp add sai -e SAI_API_KEY=sapi_YOUR_KEY -- npx -y @simular-ai/sai-mcp

Then ask your agent to use Sai for the steps that only exist on a screen, such as an admin console setting or the desktop app it just built. Setup for Codex and Cursor is in the docs.

Where Sai works

Every task runs on a computer, which the API calls a machine:

• Cloud computers. Windows or Linux computers Simular hosts for you. The free plan includes limited access to a Windows cloud computer; paid plans get their own.
• Your own computer. A Mac or Windows PC with the Sai desktop app installed. Only AI usage counts there, with no computer time.