Intro

Atmos - Atmosphere for Agentic Builders

Atmos is a development environment for the agent era: multi-workspace isolation, persistent terminals, integrated Git workflow, Canvas, global agent chat, and more. Use it on your machine via Desktop or Local Web Runtime.

Features

• Multi-Workspace Development — Git worktree isolation for parallel agent execution across multiple environments.
• Persistent Tmux Sessions — Fault-tolerant terminal management with tmux; sessions survive interruptions and restarts.
• Built-in Lightweight Editor — File preview, inline editing, and switch back to manual coding mode anytime.
• Integrated Git Workflow — Diff view, commit assistance, code review, and GitHub PR management in one place.
• Skill Management System — Discover, enable/disable, and delete agent skills with one-click control.
• Global Agent Chat Panel — Start non-terminal conversations from anywhere, powered by ACP to reuse your Code Agent CLI.
• Global Search & Command Palette — Keyboard-driven workflow for searching and executing Atmos features.
• Usage Analytics Dashboard — Track AI coding subscription quotas, agent token consumption, and cost estimation.
• Agent Status Notifications — Hook-based status monitoring with native notifications and self-hosted push server support.
• Cross-Platform & Remote Access — Desktop and browser clients; Remote Access (Desktop) shares this machine via Tailscale / Cloudflare Tunnel / ngrok and a Public URL—separate from Atmos Computer.
• Atmos Computer — Connect to a remote server through Atmos Relay (no tunnel on the client).
• Kanban View - Quickly manage Workspace status, priority, labels and other information in Kanban view.
• Canvas — A cross-project infinite canvas: pin terminal cards from any workspace or project onto one persistent board, and let Code Agents drive the canvas diagrams, notes, and layout without leaving your agent workflow.
• Review Workflow — Review changes in Atmos's built-in diff UI, leave inline comments on specific lines, then hand off to your Code Agent to apply fixes.
• Agent Status Tracking — Real-time agent lifecycle sync via hooks (running, idle, waiting for permission, done) across the UI, with notifications on state changes—native alerts plus self-hosted push (ntfy, Gotify, or a custom webhook).
• Lightweight Local Models — One-click run llama-server for small Hugging Face models on your machine—ideal for light tasks such as session titles, workspace TODO extraction, and Git commit message generation without configuring a cloud API.

In this docs