Shepherd Terminal

Product Introduction

1. Definition: Shepherd Terminal is a specialized, agent-native integrated development environment (IDE) and terminal multiplexer designed for AI-powered coding workflows. It functions as a persistent workspace where multiple AI agents, such as OpenAI Codex and Anthropic Claude, can operate concurrently across distributed systems.
2. Core Value Proposition: It exists to solve the context fragmentation and management overhead inherent in using multiple AI coding assistants. Shepherd provides a unified, stateful environment where agents can run persistently, their work can be monitored in real-time, and developers can seamlessly resume complex, multi-step coding sessions without losing terminal state, agent context, or file changes.

Main Features

1. Persistent, Survivable Terminal Sessions: Unlike standard terminals, Shepherd decouples the user interface from the underlying processes. When the application is closed, all terminal sessions (including running servers, tests, or agent commands) continue to execute in the background. Reopening the app restores the exact tab and pane layout with the live processes, enabling true long-running agent tasks and remote development without interruption.
2. Real-Time Multi-Agent Monitor: The interface includes a dedicated, always-visible agent status dashboard. This monitor displays every active AI agent (e.g., Codex, Claude Code), its current state ("working," "idle," "done"), the task it's performing, and its working directory. This provides at-a-glance visibility into parallel AI-driven development work across different tabs, panes, and even remote machines.
3. Agent-Centric Workspace Management with Plugins: Shepherd equips AI agents with a dedicated plugin API that grants them operational control over the workspace. Agents can programmatically create new terminal tabs, split panes, send input to specific sessions, and read output. This allows an agent to architect its own workspace—for example, spawning a sub-agent in one pane to write tests while it runs the test suite in an adjacent pane—without manual developer intervention.
4. Integrated Worktree and Diff Viewer for Agent Output: The platform includes a dedicated "Worktree" panel that tracks all file changes made by agents within a session. Developers can instantly review the exact diffs generated by an agent, browse modified files, and resume an agent's work context with a single click, bridging the gap between AI-generated code and human review.
5. Seamless Remote Development Integration: Shepherd treats remote SSH machines as first-class citizens. It allows developers to connect to remote servers, browse and edit remote files as if they were local, and automatically forwards remote services (like a locally-bound development server on port 3000) to the developer's local machine via secure "Routes." This creates a cohesive hybrid environment for GPU-based or cloud development.

Problems Solved

1. Pain Point: Context loss and session fragility when switching between AI agents, closing apps, or working remotely. Traditional terminals kill processes on exit, and managing multiple AI agent outputs across disparate windows is chaotic.
2. Target Audience: Software engineers and developers leveraging multiple AI coding assistants (Codex, Claude, etc.) for complex refactoring, feature development, or testing; teams engaged in remote development on cloud instances or GPU servers; developers who require persistent, monitorable long-running processes.
3. Use Cases: Running a Codex agent to refactor an authentication module while a Claude agent drafts documentation in parallel; deploying a long-running model training script on a remote GPU server and needing to disconnect/reconnect without stopping it; reviewing and merging a series of code changes generated overnight by an autonomous agent; having an AI agent set up a complete test environment by creating multiple terminal panes for the test runner, linter, and development server.

Unique Advantages

1. Differentiation: Unlike standard terminal emulators (iTerm2, Warp) or multiplexers (tmux, screen), Shepherd is built from the ground up for AI-agent collaboration, with native state persistence and agent control. Compared to generic IDEs, its core innovation is the deep integration of the agent's operational layer directly into the workspace management.
2. Key Innovation: The Shepherd Plugin API is the key innovation. It transforms AI agents from passive code generators into active workspace participants capable of manipulating the development environment itself. This, combined with the process survivability layer and the unified agent monitor, creates a feedback loop where agents can perceive, act upon, and verify their work within a stable, observable environment.

Frequently Asked Questions (FAQ)

1. How does Shepherd Terminal keep processes running after the app closes? Shepherd uses a client-service architecture where a lightweight background service (daemon) manages all terminal processes independently of the graphical user interface. When you close the app, only the UI disconnects; the service continues executing all commands and sessions, allowing for seamless reconnection and state restoration.
2. Can I use Shepherd Terminal for remote development on a Linux server? Yes, remote development is a core feature. Shepherd establishes a secure SSH connection to remote machines, allowing you to manage remote terminal sessions, browse the remote filesystem natively within the app, and automatically forward remote network services (like web servers on localhost) back to your local Mac for access.
3. What AI coding agents are compatible with Shepherd Terminal? The platform is designed to be agent-agnostic but showcases deep integration with OpenAI Codex and Anthropic Claude through its plugin system. Any AI agent or CLI tool that can utilize the Shepherd Plugin API to send commands can control the workspace, meaning support can be extended to other AI models and custom automation scripts.
4. Is Shepherd Terminal a replacement for my current IDE or terminal? It is designed as a complementary, specialized tool. It is not a full-featured IDE like VS Code for editing but is a superior terminal and agent orchestration layer. Developers often use Shepherd to manage AI agent workflows and long-running processes while using their primary IDE for focused editing, connected via Shepherd's file browser.
5. How does the agent monitoring feature work? The Agent Monitor reads status from the Shepherd service and plugin communications. It displays real-time information provided by the agents themselves (task name, status, location) and system data, aggregating activity from all connected local and remote sessions into a single, unified dashboard.