atlas

Product Introduction

1. Definition: Atlas is a native macOS desktop application (not Electron-based) that functions as an integrated development environment (IDE) and collaborative workspace specifically engineered for AI-augmented software development. It is built on the Tauri framework (Rust + WebKit) and integrates the Agent Client Protocol (ACP) to unify human planning with multiple AI coding agents like Claude Code and Codex in a single, local-first interface.
2. Core Value Proposition: Atlas solves the critical problem of workflow fragmentation in modern AI-assisted coding by providing a unified planning surface. Its core value is enabling engineering teams and individual developers to seamlessly manage the entire coding lifecycle—research, planning, knowledge management, and AI agent execution—within one performant, context-aware application, thereby tightening the feedback loop between human intent and AI-generated commits.

Main Features

1. Unified Workspace & Multi-Agent Concurrency: Atlas consolidates chat interfaces, a CodeMirror 6-powered code editor, a built-in WebKit browser, integrated terminal (zsh/bash/fish), and a spatial canvas into one native window. It supports running multiple AI agents (e.g., Claude Code, Codex) concurrently in separate session workers, allowing parallel tasks like refactoring, testing, and research across different workspaces without tab-switching or context loss.
2. Local-First Knowledge Base & Reference System: Project knowledge is stored as plain Markdown files in a .atlas/knowledge/ directory, versioned with the codebase. Atlas indexes this content, enabling a powerful @-mention reference system. Users and agents can pull in specific context by referencing files (@src/auth.ts), line ranges (#L40-80), branches, commits, knowledge notes, or even prior agent threads, ensuring precise context injection into prompts.
3. ACP-Native Agent Integration & Activity Transparency: Atlas uses the open Agent Client Protocol (ACP) as its primary integration layer for AI agents, avoiding vendor lock-in. It provides deep observability: a real-time Activity Log (JSONL on disk) records all events; an Inbox (⌘⇧I) surfaces permission prompts and agent actions; and a live Token Transparency counter displays per-thread token usage, API calls, and cost estimates to prevent surprise bills.
4. Advanced Codebase Intelligence Tools: The platform includes specialized developer tools for pre-commit analysis. The Blast Radius Analysis uses tree-sitter to show which files, symbols, and tests will be affected by a change before an agent commits code. The Complexity Heatmap ranks repository files by symbol density to identify refactoring hotspots, aiding in intentional code maintenance.
5. Spatial Canvas for Planning & Git Graph Visualization: The Spaces feature offers an infinite ReactFlow-powered canvas for creating, linking, and grouping visual notes, which can be dragged into chat as agent context. The Git Graph provides a lane-aware, visual commit history that accurately renders branch and merge topology, allowing inline diff viewing without leaving the chat thread.

Problems Solved

1. Pain Point: Context Switching and Tool Fragmentation. Developers using AI coding assistants typically juggle between a code editor, terminal, browser, documentation, note-taking apps, and separate AI chat interfaces, leading to cognitive overhead and broken workflows.
2. Target Audience: The primary users are software engineers, engineering teams, and tech leads who regularly utilize AI coding assistants (Claude Code, GPT-4, etc.) for substantive work and need to coordinate that work within a project's full context. It is especially valuable for developers working on complex refactors, planning new features, or onboarding to large codebases.
3. Use Cases:
   • Coordinated Multi-Agent Workflows: Running one agent to write tests while another refactors a related module, with both having shared access to the same knowledge base and git context.
   • Onboarding & Codebase Exploration: Using the spatial canvas to map out architecture, the @-reference system to pull relevant code, and AI agents to explain complex sections, all within one window.
   • Pre-Commit Impact Analysis: Using Blast Radius and Complexity Heatmap tools to audit an AI agent's proposed changes for unintended consequences before merging.
   • Persistent Project Knowledge Management: Maintaining a living CLAUDE.md or AGENTS.md and project notes in the .atlas/ directory that both humans and AI agents can read and update, creating a single source of truth.

Unique Advantages

1. Differentiation: Unlike generic code editors with AI plugins or standalone AI chat interfaces, Atlas is architected from the ground up as a planning surface for coding agents. It deeply integrates project context, agent memory, and development tools, whereas competitors treat AI as a side feature. Its local-first, file-based knowledge system contrasts with cloud-only, proprietary agent memory systems.
2. Key Innovation: The Agent Client Protocol (ACP) as a core architectural primitive allows for true multi-agent, multi-vendor concurrency in a stable, interoperable way. Furthermore, its state management model places all business logic in Rust (via Tauri), with React serving only as a view layer. This ensures performance, state persistence across window reloads, and a single source of truth, differentiating it from typical Electron apps.

Frequently Asked Questions (FAQ)

1. Is Atlas a cloud-based AI tool or a local application? Atlas is a native, local-first macOS desktop application. Your code, project knowledge (.atlas/ directory), and activity logs reside on your machine. AI API calls (e.g., to Anthropic, OpenAI) are made directly from your system using your own API keys, ensuring data privacy and control.
2. How does Atlas handle collaboration and team use? The current alpha version is optimized for single-player use. However, its architecture is designed for teams. The roadmap includes features like Atlas Accounts for sync, shared team knowledge bases, thread synchronization, and version control for agent context, aiming to make AI agent collaboration seamless across engineering teams.
3. What AI models and coding agents does Atlas support? Atlas natively supports any AI agent that implements the open Agent Client Protocol (ACP). It ships with built-in integrations for Claude Code and Codex via ACP. Support for other ACP-compliant agents like OpenCode and Kilo Code is planned. This protocol-based approach prevents vendor lock-in.
4. How does the "local-first" knowledge base work with AI agents? Atlas stores project notes and knowledge as standard Markdown files in a .atlas/knowledge/ folder within your repository. This directory is indexed locally. When you or an agent uses the @knowledge/note-name reference, Atlas reads the file directly from disk and injects its content into the prompt. This keeps knowledge versioned with code and accessible offline.
5. What are the system requirements for running Atlas? Atlas requires macOS 13 (Ventura) or later and runs natively on both Apple Silicon (ARM) and Intel-based Macs. It is a compiled Rust application via the Tauri framework, resulting in a compact, high-performance binary without the overhead of Electron, making it suitable for long development sessions.