# Shelter41

> Infrastructure platform for running long-running AI coding agents in persistent, pausable cloud sandboxes with orchestration, observability and governance.

Shelter41 provides infrastructure for long-running AI agents. It offers persistent cloud workspaces for coding agents such as Claude Code and Codex, with browser, CLI and SDK access, and enterprise options that are managed by Shelter41 or run on a customer's own Kubernetes cluster.

## What It Is

Shelter41 is an agent orchestration and workspace platform. A user describes a task, such as building, testing and reviewing a sign-in flow, and an orchestrator chooses agents, allocates an isolated sandbox for each, and pauses or resumes compute as work proceeds. The site's example sequence is labeled illustrative.

## Persistent Workspaces and Idle Compute

Shelter41 separates an agent's identity, session history and workspace memory from its compute. Compatible sandboxes sleep automatically after a configurable idle delay, including supported waits for user input, and wake when new work arrives. Pausing keeps files and the running environment; parking removes compute while retaining agent identity and recorded history. Agents in one workspace can share project observations through workspace memory, while each runs in its own sandbox. Per the FAQ, recovering a native conversation after a sandbox is destroyed depends on the harness, and the documented path currently covers Claude Code.

## Control, Observability and Governance

The platform's control view tracks running, waiting and paused agents. Observability covers run reliability, performance, token usage and cost by runtime, filterable by workspace, runtime, agent or user. Governance covers organization roles, per-member access to runtime credentials, and trace and tool-payload capture settings. A skills library lets teams install skills on compatible agents. The dashboard figures shown are sample data.

## Integrations and Deployment

Listed integrations include GitHub, GitLab, Linear and Supabase via MCP, plus the sh41 CLI. Slack, WhatsApp and Telegram are marked coming soon. The FAQ says the built-in harness catalog includes Claude Code, Codex, Kimi Code and experimental Grok Build; other harnesses shown illustrate the broader concept. Enterprise customers can choose a managed deployment or run on their existing Kubernetes infrastructure; fully offline requirements need separate review.

## Features
- Agent orchestration across multiple coding agents
- Isolated sandbox per agent
- Automatic sandbox pause and resume with files and context retained
- Persistent workspaces with shared project memory
- Observability of runs, token usage and cost
- Governance of roles, runtime credentials and trace capture
- Shared skills library for compatible agents
- sh41 CLI and SDK
- Managed or on-premises Kubernetes enterprise deployment

## Integrations
Claude Code, Codex, Kimi Code, Grok Build, GitHub, GitLab, Linear, Supabase, Kubernetes

## Platforms
WEB, API, DEVELOPER_SDK, CLI

## Pricing
Paid

## Links
- Website: https://shelter41.com
- EveryDev.ai: https://www.everydev.ai/tools/shelter41