Skip to main content
arg.ai gives your code and your agents the same workspace: durable files, a sandbox to run commands in, and an agent you can point at both over a REST API.

Quickstart

Create a workspace, upload a file, run an agent, collect the output - in curl.

Agents

Start a run, watch it, steer it, and stop it. The heart of the API.

Start here

Authentication

API keys, and what each kind of credential can reach.

TypeScript SDK

The same flow in a few typed calls.

Command line

Move files, install skills, mount a workspace into your coding agent.

Build with agents

Agent runs

The run lifecycle: start, poll, queue, steer, stop.

Skills and subagents

Teach a workspace how you want recurring work done.

Unattended runs

Service accounts, schedules, and knowing when something broke.

MCP

Connect Claude Code, Codex, or any MCP client to a workspace.

Build apps

Arg SDK

One import for files, chats, Actions and resources - inside Arg, on a server, or in an agent’s own scripts.

React app SDKs

Build live TSX and JSX apps with shared UI and typed Actions.

Full-view apps

Open finished HTML, TSX, and JSX apps as preview-first experiences.

Core concepts

Organizations

Teams, roles, invitations, and shared workspaces.

Workspaces

Cloud-backed volumes for files and sandboxed execution.

Files

Upload, download, read, write, version, and share.

Tools

Run commands and edit files directly, with no model in the loop.

Actions

A typed catalog of capabilities your code and your agents both call.

Search

Semantic search across a workspace or a whole organization.

Servers

Expose a process from the sandbox on an HTTPS URL.

Integrations

Connect the accounts your agents act on.

API Reference

Explore the API

Every endpoint with request and response schemas, runnable samples, and an interactive playground.