One Roof

Product Introduction

1. Definition: One Roof is a macOS-native, open-source desktop application designed for multi-agent software development orchestration. It functions as a centralized command center, allowing developers to concurrently run, monitor, and manage multiple AI-powered coding agents (like Claude Code, Codex, Gemini CLI, Kimi, and Antigravity) within isolated, persistent terminal sessions.
2. Core Value Proposition: It exists to solve the operational complexity and cognitive overload of "chorus coding"—using several specialized AI coding agents collaboratively on a single project. Its primary value is maintaining developer oversight and control, providing a unified interface for agent output, questions, and system resource management, thereby increasing the efficiency and safety of AI-augmented development workflows.

Main Features

1. Unified Agent Inbox (⌘I): A centralized, priority-sorted queue for all agent-initiated queries and required approvals. Technically, it intercepts and aggregates prompts from all active terminal sessions. Each request block presents the agent's own suggested action buttons (1-9), and users can reply with text or paste screenshots for visual context. This feature ensures no agent remains blocked in an unattended terminal tab.
2. Persistent Session Management via roofd: Sessions are not simple terminal emulators but are managed by roofd, a dedicated background daemon. This architecture allows agent sessions to remain alive and processing even when the main One Roof application is closed or updated. Sessions are visually categorized by state (Needs You, Active, Next, etc.), display live status, and can be viewed in single, grid, or wall layouts.
3. Automated Loop Execution with Guardrails: This feature enables scheduled or conditional repeated agent runs (e.g., "run tests every night"). Loops operate on a separate branch or copy of the repository, not the main working directory. The system runs a check first, only triggering an agent if the condition fails. All results are staged for manual review in the Inbox; loops never autonomously commit, push, or merge code, enforcing a key safety boundary.
4. System-Wide Resource Dashboard: Provides real-time and historical metrics on CPU and memory usage per agent session, per application, and for the system overall. It includes a built-in process guard that warns when a session exceeds 1GB of memory and terminates it at 2GB to prevent system slowdowns or freezes, addressing a critical pain point in multi-agent environments.
5. Contextual History & Morning Briefing (Home): The History feature condenses session activity into a searchable, card-based timeline of "important moments" (steps, errors, hand-offs) for the past 30 days. The Home screen synthesizes this data into a morning briefing, highlighting overnight loop failures, pending actions ("Decide now"), and a summarized timeline, enabling rapid project status assessment.

Problems Solved

1. Pain Point: Context fragmentation and lost agent queries when managing multiple concurrent AI coding sessions across disparate terminal windows and tabs.
2. Pain Point: Lack of visibility into autonomous or overnight agent activity, leading to missed failures, stalled progress, and difficulty resuming work.
3. Pain Point: Uncontrolled resource consumption by individual agent processes, which can degrade overall system performance without clear attribution.
4. Target Audience: Senior Software Engineers, Tech Leads, and AI Engineering specialists who actively integrate multiple AI coding assistants into their daily development, code review, and testing pipelines. It is particularly valuable for developers practicing "agentic" or "multi-agent" workflows on complex projects.
5. Use Cases: Orchestrating a code review where one agent (Claude) writes implementation, another (Codex) reviews for security flaws, and a third (Kimi) runs unit tests. Managing a continuous integration loop that attempts to fix broken tests overnight. Providing a safe, auditable environment for junior developers to experiment with powerful AI coding tools under supervised constraints.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):
   • Pros: Exceptional focus on developer control and observability. Vendor-neutral design avoids lock-in. The persistent roofd service is a robust architectural choice. Safety-first defaults (Manual mode, no auto-merge) are critical for professional use. The integrated resource monitor is unique and practical.
   • Cons: Platform-locked to macOS 14+ on Apple Silicon, excluding Windows, Linux, and Intel Mac users. Requires pre-installed and authenticated CLI tools for each agent, adding setup complexity. The "Evals" and "Learn" modules are marked as drafts, indicating incomplete core features. Relies on community support (GitHub) as a free, open-source project.
2. Key Alternatives & Differentiation:
   • VS Individual Vendor IDEs (Cursor, Windsurf, Zed with AI): These are editor-centric and typically optimized for a single vendor's AI model. One Roof is agent-agnostic, terminal-centric, and built for orchestration across vendors, not deep editing integration within one.
   • VS DIY Terminal Multiplexers (tmux, screen): While tmux can manage multiple sessions, it provides zero semantic understanding of agent state, no unified query inbox, no resource guards, and no automated loop scheduling. One Roof adds a high-level management layer specifically for AI agent workflows.
   • VS Cloud-Based AI Platforms (Replit, GitHub Copilot Workspace): These are closed, cloud-hosted environments. One Roof is a local, open-source desktop application that runs on your hardware with your existing local toolchain and API subscriptions, offering greater privacy, cost control, and integration with local dev environments.

Frequently Asked Questions (FAQ)

1. Is One Roof a replacement for Claude Code or GitHub Copilot? No, One Roof is an orchestration and management layer. It does not replace AI coding agents; it runs their official command-line interfaces (CLI) in managed sessions. You must have and pay for the agents (e.g., Claude Pro, GitHub Copilot) separately.
2. How does One Roof handle my API keys and security? One Roof follows a "no keys in the app" principle. It leverages the existing authentication of the installed agent CLIs (e.g., claude auth). For optional features like Linear integration or hosted models, you provide your own API keys directly to those services. The app is code-signed, notarized by Apple, and its source is publicly auditable on GitHub.
3. Can I use One Roof for free, and what are the costs? The One Roof application itself is free and open-source (MIT license). However, you incur the standard usage costs of the underlying AI agents you run through it (e.g., Claude Pro subscription, OpenAI API costs for Codex, Google AI Studio costs for Gemini). There is no additional fee for One Roof.
4. What is the "roof kit" and do I need it? The roof kit is an optional command-line toolset that enforces workspace discipline. It installs shell hooks and agent wrappers to ensure all AI sessions automatically start in your designated project directory and read common rule files (like CLAUDE.md). It's not required to run the app but is recommended for consistent multi-agent context.
5. Does One Roof work with any AI model or just the listed ones? It primarily supports the listed agents with dedicated integration (Claude Code, Codex, etc.). However, it can also run "Plain shells," meaning you could theoretically run any CLI-based AI tool within a session, though you may miss some agent-specific UI integrations.