Puppet Master unifies multiple AI coding agents into a single dashboard

Puppet Master v0.10.0 has launched as a unified interface for managing multiple AI coding agents simultaneously. Released on October 7, 2026, this tool allows developers to spawn and supervise real sessions of Claude Code, Codex, Gemini, and other agent CLIs from a single dashboard. It runs on macOS and Linux, providing a centralized view of agent activity across all open projects.

What happened

The release introduces a daemon-based architecture that oversees independent agent sessions without replacing the underlying CLI tools. Instead of reimplementing agent logic, Puppet Master starts the standard command-line interfaces developers already use and monitors their output. This approach ensures that prompts, permission requests, and terminal behavior remain identical to native usage, preserving existing workflows while adding oversight capabilities.

Previously, running multiple agents required maintaining several terminal windows and mentally tracking the state of each session. This manual management often led to lost context, such as missing when an agent finished a task or drifted off course. Puppet Master addresses this scaling problem by owning the pseudo-terminals (PTYs) and reporting coarse state changes like working, idle, or needs input. The system uses an Model Context Protocol (MCP) channel wired into every session to report progress directly from the agent, rather than guessing based on scraped terminal text.

How it works

The core mechanism relies on workers, which are machines that execute the sessions assigned by the controller. By default, the controller’s local machine acts as a worker, but users can enroll additional machines, virtual machines, or containers as separate workers. This allows for flexible resource allocation, such as isolating noisy jobs on specific hardware or sharing a single machine among multiple agents. Sessions are placed on specific workers either by user choice or by project configuration, ensuring consistent file paths and environments.

Security and isolation are handled through sandboxed workers. Users can start a worker in a container using Incus, Docker, or Podman, mounting only specific host directories. This setup prevents agents from accessing the entire host filesystem while allowing them to operate within a disposable environment. The controller manages credentials for external connections, such as MCP servers or REST APIs, ensuring that keys never appear in agent conversations. Reads typically run automatically, while writes require explicit user approval via the dashboard, with every action logged for audit purposes.

Key details

• Multi-platform support: Runs on macOS and Linux with support for various agent CLIs including Claude Code and Codex.
• Live terminal access: Sessions are rendered with xterm.js, allowing typing, resizing, and scrolling from the browser, iOS app, or TUI.
• Sandboxed execution: Workers can run in containers with restricted filesystem access and configurable CPU and memory limits.
• Secure credential management: The controller stores API keys and handles downstream requests, keeping secrets away from agent prompts.
• Cross-device synchronization: An iOS app provides full terminal access and push notifications for sessions requiring user input.
• Persistent state: Sessions outlive shell closures, with state persisted to SQLite for later reattachment.

Why it matters

For software engineers building with AI, the ability to run multiple agents concurrently is increasingly valuable but operationally complex. Puppet Master reduces the cognitive load of managing these parallel tasks by providing a single source of truth for session status. Developers can focus on one agent while others continue working in the background, confident that they will be notified immediately if input is required. This efficiency gain is critical for teams looking to scale their AI-assisted development without sacrificing control or visibility.

The security model also addresses a significant concern in AI integration: credential leakage. By intercepting API calls and requiring approval for write operations, the tool prevents agents from accidentally exposing sensitive data or making unauthorized changes. The detailed audit log of every tool call, including arguments and results, provides necessary transparency for compliance and debugging. This structured approach to agent supervision makes it safer to delegate more complex tasks to AI systems.

What you can do

• Install the tool on macOS or Linux using the provided curl command and start the daemon.
• Configure a sandboxed worker with limited directory access to test agent behavior safely.
• Set up connections for external APIs, defining read and write policies to control agent permissions.
• Use the iOS app to monitor sessions and respond to approval requests while away from the desk.
• Create instruction overlays for projects to enforce coding conventions across all spawned sessions.
• Review the documentation to understand how to enroll multiple workers for distributed processing.