Get started free
Documentation

intentic documentation

intentic gives each coding agent its own sandbox — a specialized workspace on hardware you own. These docs cover getting one running, how the system fits together, and how to turn a sandbox into an autonomous employee.

The mental model

One idea holds the whole product together: a specialized agent is a purpose-built sandbox. Not a prompt or a markdown file — a real container, on a machine you control, with the libraries, dev-tools, system access, and curated context a specific job needs. You drive it from any browser; it runs where you started it.

Two tiers make that work, and only your sandbox holds your code:

  • The platform — a thin hub that stores your identity and your sandbox's URL. It sits off the command path and can't reach in.
  • Your sandbox — one container per agent, on your hardware, reached by your browser over a private Cloudflare tunnel. The agent, repos, credentials, and capabilities live here, along with any tool it runs — a database client, a headless browser, even a deployment engine.

Read Architecture for the full picture, or jump straight to bringing one up.

Start here

Quickstart

Sign in, name a sandbox, and paste one curl command to bring a live agent workspace up on your own machine. No inbound ports.

Docker setup

What the sandbox container is, how Docker is installed, the isolated in-sandbox engine, environment overlays, and the update/rebuild/cleanup helpers.

Architecture

How intentic fits together — the thin platform, the per-agent sandbox on your hardware, the ownership and trust model, and the Cloudflare reachability fabric.

Autonomous employees

Specialize a sandbox into an agent — its tools, systems, and context — give it work, make it event-driven, then scale to a team of them.

Reference architecture

An example of an entire company assembled from intentic sandboxes — one co-piloted agent per role and team, connected to the services they share. Every block maps to a shipping primitive.

What's in a sandbox

Everything an agent needs to do real work, and nothing the platform can see:

  • An agent — Claude Code, Codex, or Grok, on your own subscription, chosen per conversation.
  • A workspace — your repos, a file tree, editors, and terminals that survive reconnects.
  • Capabilities — GitHub, databases, Sentry, Stripe, SSH hosts, MCP servers, added in a click.
  • An environment — libraries and dev-tools baked into the image, proposed by the agent and applied on your approval.
  • Automations — schedules, webhooks, and event listeners that wake the agent while you're away.
Next Quickstart →