Yardlet

Product Introduction

1. Definition: Yardlet is a free, open-source, native macOS application (written in Swift) that functions as a unified interface and orchestration layer for multiple AI-powered coding agents. It is not a web wrapper but a desktop application that integrates directly with the command-line interfaces (CLIs) of existing coding assistants.
2. Core Value Proposition: Yardlet exists to eliminate context switching and workflow fragmentation for developers who use multiple AI coding agents like Codex, Claude Code, and OpenCode. Its primary value is providing a single, cohesive workspace where these agents can run side-by-side, interoperate, and be managed efficiently without altering their core functionality or data stores.

Main Features

1. Unified Agent Workspace: Yardlet launches individual agent CLIs (e.g., codex, claude) in parallel, presenting them through a consistent user interface comprising a shared sidebar, conversation view, composer, and approval cards. Crucially, it does not proxy or intercept sessions; each agent maintains its own local session store (e.g., CODEX_HOME), ensuring compatibility with their native apps.
2. Cross-Agent Orchestration Tools: Yardlet equips every integrated agent with a suite of session-level tools for inter-agent communication and task management. Key tools include read_task, send_message_to_task, create_task (to spawn child tasks on any agent), run_agent (for one-off subagent execution), and create_scheduled_task. This enables complex, multi-agent workflows like having Claude Code draft code and Codex review it.
3. Integrated Development Environment (IDE) Features: Beyond agent management, Yardlet includes developer-centric tools adjacent to the conversation pane. This includes a diff viewer for changes from the last turn, file restoration, a terminal, a files pane, and a dedicated pull request card that integrates with the gh CLI to track CI checks, display logs, and merge PRs.
4. Agent-Drivable Browser with Secure Vault: A built-in Chromium panel can be controlled by agents using browser automation tools. A key security feature is a local vault that allows agents to fill saved usernames/passwords, 2FA codes, and passkeys into sites without exposing the credentials to the AI model itself. The browser supports hand-off for human-only steps like CAPTCHAs.
5. MCP (Model Context Protocol) Server Integration: Yardlet simplifies connecting agents to external data and services via a built-in MCP server for Google services (Gmail, Calendar, Drive) and support for remote MCP servers (e.g., GitHub, Slack). Connectors are set up once and made available to all agents, enhancing their context and capabilities without manual configuration per task.
6. Multi-Account Management & Limit Visibility: The interface displays usage limits (session and weekly) for each connected account across providers. When starting tasks, Yardlet can automatically select the account with the most remaining quota, and parallel tasks are distributed across accounts to optimize limit utilization.

Problems Solved

1. Pain Point: Workflow Fragmentation and Context Switching. Developers using multiple AI coding assistants must constantly switch between different apps, terminals, or browser tabs, losing focus and breaking their flow. Managing separate session histories and approval UIs for each agent is inefficient.
2. Target Audience: Professional Software Engineers and Developers who actively use and pay for multiple AI coding agents (e.g., Codex, Claude Code) and seek to maximize their productivity through orchestration. Secondary audiences include technical leads and open-source contributors who value transparency and control offered by open-source, local-first tooling.
3. Use Cases:
   • Multi-Agent Code Review: Starting a task in Claude Code to implement a feature, then using Yardlet's run_agent tool to have Codex perform a security-focused review of the generated code within the same session.
   • Cross-Service Automated Workflows: Using an agent with Gmail and Calendar connectors to parse an email about a bug, create a GitHub issue via an MCP server, draft a fix, and schedule a calendar reminder for follow-up.
   • Quota-Optimized Development: A developer with team and personal accounts for a single AI provider can let Yardlet automatically route tasks to the account with the most available capacity, effectively managing API costs and limits.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pro (Non-Invasive Architecture): Yardlet's greatest strength is its non-invasive design. It acts as a front-end and router, not a backend. Agents retain their original sign-ins, session stores, and configurations. Users can stop using Yardlet at any time with zero impact on their existing agent setups.
   • Pro (Powerful Orchestration): The built-in cross-agent communication tools (send_message_to_task, create_task) enable sophisticated, automated workflows that are impossible when agents are siloed in separate applications.
   • Pro (Privacy-Focused & Local-First): With no analytics, no Yardlet server processing user data, and end-to-end encrypted remote access, it prioritizes user privacy. Sensitive data like connector sign-ins and vault passwords are stored locally in the macOS Keychain.
   • Con (Platform Exclusivity): A significant limitation is its restriction to macOS 15+ on Apple Silicon. Developers on Windows, Linux, or Intel Macs cannot use it.
   • Con (Setup Complexity): The value proposition requires pre-installed and configured agent CLIs. It is not a standalone solution, adding a layer of setup complexity for less technical users.
   • Con (Early Development Stage): As version 0.11.x indicates, it is beta software. Features like the companion Android app are noted as not yet publicly available, and users may encounter bugs or missing polish.

2. Key Alternatives & Differentiation:

   • Using Native Agent Apps/CLIs Individually: The standard alternative is using each agent's dedicated application or CLI in separate windows. Differentiation: Yardlet eliminates constant app switching, provides a unified UI, and enables direct agent-to-agent communication, which is impossible in isolated apps.
   • Cloud-Based AI Workflow Platforms (e.g., Cursor with multi-model): Platforms like Cursor may integrate multiple models but typically do so through a single, proprietary interface and backend. Differentiation: Yardlet is agent-agnostic, open-source, and runs models through their official local CLIs, preserving user choice, direct provider relationships, and local session data. It is a coordinator, not a gateway.
   • DIY Scripting for Orchestration: A technically advanced user could script interactions between agent CLIs. Differentiation: Yardlet provides a robust, GUI-driven product with approval flows, a secure vault, a drivable browser, and MCP integration out-of-the-box, saving immense development and maintenance time.

Frequently Asked Questions (FAQ)

1. Is Yardlet safe to use with my existing Codex and Claude Code accounts? Yes. Yardlet does not require your agent credentials. It launches the official codex or claude CLIs that are already signed in on your system. Your sessions and data remain in the agents' own secure stores (like ~/.codex), and Yardlet cannot access your login tokens.
2. What happens to my work if I uninstall Yardlet? Nothing is lost. Since Yardlet does not modify agent configurations or move session data, all your conversations and project contexts remain intact within each agent's own storage. You can resume working directly in the Codex app, Claude Code app, or their respective CLIs immediately.
3. How does the cross-agent messaging and approval system work? When Agent A sends a message or approval to Agent B, Yardlet cryptographically verifies that the quoted user text exists verbatim in the original task's message history. It then forwards the message with a verifiable timestamp and a header stating it was relayed, preventing agents from misattributing statements to the user.
4. Can I use Yardlet for free, and what are the costs? The Yardlet application itself is free and open-source (MIT License). You only pay for the usage of the underlying AI agents (e.g., your Codex or Claude subscription/API costs) as you normally would. There is no Yardlet subscription or account fee.
5. Does Yardlet work with GitHub Copilot? The listed agents (Codex, Claude Code, OpenCode, etc.) are CLI-based. GitHub Copilot is primarily an IDE extension. Unless Copilot offers a full-featured, standalone CLI that Yardlet can harness, integration would not be possible. Yardlet's architecture is built around orchestrating agent processes, not IDE plugins.