Deploy SAM
All Features

Come back days later to the same session

Sessions you leave idle go to sleep and keep their work for up to seven days.

  • Sessions normally restore home and repository state from a 7-day snapshot; fallback keeps the transcript and available backups
  • Agents dispatch subtasks to peers and send durable messages that can interrupt a busy agent
  • App deployments run Compose releases with environments, configuration, and logs
  • Six agent harnesses: Claude Code, Codex, Gemini CLI, Mistral Vibe, OpenCode, Amp
Read the guide
A sleeping chat session being woken: a banner reads Recreating your workspace, the transcript is intact, and the composer reads Waking the agent, your message will be deliveredA sleeping chat session being woken: a banner reads Recreating your workspace, the transcript is intact, and the composer reads Waking the agent, your message will be delivered

Close the tab. The session waits.

SAM normally restores the agent’s home directory and repository work from a verified snapshot for up to seven days. After repeated snapshot failures, SAM releases idle compute and keeps the transcript and available backups. SAM shows a notice that unpublished files and saved agent state may be lost. Continuation uses a verified Git recovery point when one exists, or starts fresh from the transcript.

A finished chat session in the Sleeping state, with the full transcript still visible and a composer that reads Send a message to wake the agentA finished chat session in the Sleeping state, with the full transcript still visible and a composer that reads Send a message to wake the agent Sessions normally sleep after idle turns and wake from a seven-day snapshot

Agents that dispatch to each other

An agent can dispatch a subtask to a peer and send a durable message that interrupts a busy agent instead of waiting behind it. Work keeps moving even when one agent is heads-down on something else.

Task Hierarchy view showing a running parent task with three dispatched subtasks: one completed with a PR opened, one queued and blocked on sign-off, and one runningTask Hierarchy view showing a running parent task with three dispatched subtasks: one completed with a PR opened, one queued and blocked on sign-off, and one running A parent task and the subtasks it dispatched, with live status for each

Deploy what the agent built

App deployments turn a Compose release into a running app. Each environment has its own deployment node, public routes, configuration, logs, and volumes, managed from the same project — and a release can be submitted by an agent task. No separate deployment pipeline to wire up.

The staging deployment environment page: Live, running version 14 at staging.payments.northwindlabs.dev with Open app and Logs, credited to the Claude Code agent and the chat it came from; recent deploys v14, v13 and v12 with the agent and chat behind each; and who can deploy next to the settings the app readsThe staging deployment environment page: Live, running version 14 at staging.payments.northwindlabs.dev with Open app and Logs, credited to the Claude Code agent and the chat it came from; recent deploys v14, v13 and v12 with the agent and chat behind each; and who can deploy next to the settings the app reads One page per environment: what's running, which agent shipped it from which chat, and who may deploy next

Ready to try SAM?

SAM is open source. To deploy it, you need a Cloudflare Workers Paid plan, a domain on Cloudflare, and a GitHub account.