Atelier

Product Introduction

1. Definition: Atelier is a native macOS desktop application (Developer ID notarized) that functions as a centralized project management and memory system for AI-assisted software development. It integrates via the Model Context Protocol (MCP) server and custom hooks to provide context and continuity for AI coding agents like Claude Code, Codex, and Cursor.
2. Core Value Proposition: Atelier exists to extend the utility and reduce the operational cost of AI coding subscriptions (specifically Anthropic's Claude plans) by providing persistent memory, structured workflows, and automated context management. Its primary value is in making AI development sessions more efficient and coherent, directly addressing the token-waste and context-loss problems inherent in stateless AI chat interfaces.

Main Features

1. Contextual Briefing System: Atelier generates a concise, structured brief for every new AI coding session. This brief dynamically pulls from the project's current state: active tasks, recent feedback, session history from the previous agent, project-specific rules, and a "memory" database. It works by having two small Claude Code hooks automatically inject this context at the start of a session, ensuring the AI begins with relevant information without manual copy-pasting.
2. AI-Native Project Management: The app provides a unified studio for each software project, containing roadmaps, tasks, releases, feedback, and a persistent memory. AI coders can read and write to this studio via the built-in MCP server. This feature works by linking a local code folder to an Atelier project, allowing the MCP-enabled AI tool to directly query and update project data, keeping documentation and plans synchronized with code changes.
3. Cost-Optimized Code Review ("Second Opinion"): For high-risk code changes (e.g., security, payments), Atelier facilitates a two-stage review. First, the primary AI coder (e.g., Claude) requests a read-only analysis from OpenAI's Codex via the local ChatGPT desktop app. Codex returns a list of potential issues. Second, the primary coder evaluates each finding, providing a verdict and fixing only the valid ones. This process leverages a user's existing ChatGPT subscription, avoiding extra API costs while using Codex for initial scanning, which reportedly uses about 71% fewer Claude tokens.
4. Session Autopsy & Continuity: At the end of each coding session, the AI agent is prompted to write a "session page" documenting what it did, learned, was slowed by, and recommends next. This page is saved within Atelier and forms part of the brief for the next session. The "Activity" log tracks all changes, and entire sessions can be rolled back from their page, creating a reversible, auditable development history.
5. Human-in-the-Loop Approval Workflow: AI coders cannot mark tasks as complete. Instead, they submit "finished work" to the Today and Review queues within Atelier, including a summary of changes and testing instructions. A human developer must manually approve or send back the work with notes, which then become priority input for the next AI session on that project.

Problems Solved

1. Pain Point: Context Amnesia in AI Coding. Each new chat with an AI coder starts from a blank slate, forcing developers to repeatedly paste project history, rules, and incomplete work, wasting time and consuming significant context window tokens.
2. Pain Point: Unmanaged AI Development Chaos. Without a central system, notes, tasks, and decisions made with AI assistants become scattered across chat histories, notes apps, and issue trackers, leading to disorganization and lost knowledge.
3. Target Audience: Indie Mac Developers & Small Teams using AI coding tools (Claude Code, Cursor) for building and maintaining macOS, iOS, or cross-platform applications. It targets developers who are cost-conscious about their AI subscription plans and need to manage multiple projects.
4. Use Cases: Extending a limited Claude plan by offloading initial code review to Codex. Onboarding an AI to a legacy project by centralizing its rules and memory. Managing parallel feature development across multiple apps with separate AI sessions. Creating auditable trails of AI-generated code changes for debugging or compliance.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros:
     • Direct Cost Savings: Quantifiably reduces Claude token usage via the Codex-assisted review and brief-focused sessions, directly extending the value of a subscription plan.
     • Native macOS Integration: Runs as a notarized desktop app with optional iCloud sync and Menu Bar extra, offering a seamless, private experience without web accounts.
     • Open Integration via MCP: While optimized for Claude Code, its built-in MCP server makes it compatible with any MCP-capable tool (Cursor, Windsurf, etc.), avoiding vendor lock-in.
     • One-Time Purchase Model: Unlike many SaaS tools, it's a single payment with free updates, favorable for long-term use.
   • Cons:
     • Platform Lock-in: Exclusively for macOS 14+, excluding Windows, Linux, and web-based developers.
     • Dependency on Other Apps: Full functionality requires Claude Code for briefs/session pages and the ChatGPT app for the "second opinion" feature.
     • Learning Curve: Requires setup of MCP connections and Claude Code hooks, which may deter non-technical users.
     • Niche Focus: Primarily valuable for developers heavily invested in the specific workflow of AI-assisted coding with Claude and Codex.

2. Key Alternatives & Differentiation:

   • vs. Generic Project Management (Linear, Jira): Atelier is not a general-purpose PM tool. Its differentiation is bi-directional, automated integration with AI coders. While Linear manages tasks for humans, Atelier's data is designed to be read and written by AI agents, with features like automatic brief generation and session logging that generic tools lack.
   • vs. AI-Powered Development Agents (Smithery, Aider): These tools are primarily code-centric agents. Atelier differentiates by being a context and workflow manager that sits alongside these agents. It doesn't replace Claude Code; it makes it more efficient by managing its memory and workflow. It's a coordinator rather than a competitor.
   • vs. Manual Context Management (Notion, Craft): Using a manual wiki requires constant copy-pasting and lacks automation. Atelier's key difference is automated context injection and session logging via MCP, creating a living document updated by the AI as it works, which static note-taking apps cannot do.

Frequently Asked Questions (FAQ)

1. Does Atelier work with GitHub Copilot or VS Code's built-in AI? Atelier works with any development tool that supports the Model Context Protocol (MCP). This includes Cursor and any VS Code setup with an MCP client extension. It does not directly integrate with GitHub Copilot's proprietary chat interface, but if Copilot is used within an MCP-enabled editor like Cursor, Atelier's project context could be accessible.
2. Is my code or project data sent to Atelier's servers? No. All data, including project memory, tasks, and session logs, is stored locally on your Mac. Optional iCloud sync uses your personal, private iCloud account. The developer states the app has no analytics, tracking, or external data transmission, aligning with a privacy-focused, local-first architecture.
3. How does the "71% fewer Claude tokens" claim work in practice? The savings come from a division of labor. Instead of using the more expensive Claude model to both find potential issues in a code diff and then judge them, Atelier's workflow uses Codex (via your local ChatGPT app) to perform the initial "finding" scan read-only. Claude then only receives the list of findings to judge and act upon, significantly reducing the volume of code it needs to process in its context window.
4. Can I use Atelier for team collaboration? The current implementation is fundamentally single-user. While iCloud sync can share the studio across a user's multiple Macs, it is not designed for multi-user concurrent access or role-based permissions. Its workflow centers on an individual developer managing their AI coding assistants.
5. What happens after the 7-day free trial? After the trial, the app requires a one-time purchase license to continue functioning. The license is valid for the major version (e.g., 1.x) with free updates and can be activated on up to 3 Macs owned by the licensee. No subscription or ongoing payment is required.