# GooeyPi: One Desktop Shell for Pi, OMP, and Prime Agent > GooeyPi is a cross-platform desktop workspace that wraps three coding-agent harnesses in a single GUI. It leaves models, logins, and sessions with each harness while adding a unified interface, scheduling, and browser control. **am-will/gooey-pi** — Desktop workspace for Pi, OMP, and Prime Agent - Repository: https://github.com/am-will/gooey-pi - Website: https://github.com/am-will/gooey-pi - Stars: 950 · Forks: 79 - Language: TypeScript - License: MIT - Published: 2026-08-26 · Updated: 2026-08-26 · Language: en - Canonical page: https://hysenlabs.com/projects/am-will-gooey-pi ## The problem: three CLIs, one screen Pi, OMP, and Prime Agent are separate coding-agent harnesses, each with its own models, logins, and saved sessions. Developers who use more than one of them must switch between different terminal workflows and remember which commands belong to which tool. GooeyPi addresses this by providing a single desktop application that wraps all three. The README positions it as a workspace, not a replacement: each harness keeps ownership of its own models, provider credentials, and session storage. This is a meaningful design decision. It means GooeyPi does not unify your data; it unifies the interface you use to reach that data. The target user is someone who already has one or more of these harnesses installed and wants a GUI with project organization, scheduling, and browser control without abandoning the underlying CLI tools. ## How the harness switcher works The core mechanism is a harness switcher in the top-left corner. GooeyPi detects which harnesses are installed on the machine and displays only the ones that are ready. The README says you can refresh detection or set a custom executable path in Settings → Harness. Each harness has its own workspace state: projects, sessions, model lists, provider visibility, and running work. Switching harnesses changes the workspace but does not stop work already running in another harness. This implies that GooeyPi runs each harness as a separate process or at least keeps independent state per harness. The documentation does not specify the exact process model, but the behavior is clear: you can have an OMP session running and switch to Prime Work without interrupting it. This is a practical approach for users who run long agent tasks and want to check on another harness in parallel. ## Getting it running: source and releases The simplest path is to download from GitHub Releases. The README lists a DMG and ZIP for macOS, common Linux package formats plus AppImage, and a Windows installer and ZIP. For running from source, Node.js 24.15.0 or newer and npm 12.0.2 or newer are required. The repository pins Node.js 24.15.0 in `.nvmrc`. With nvm, the commands are `nvm install && nvm use`, then `npm run toolchain:bootstrap`. The README is truncated at that point, so the exact subsequent steps are not shown. That is a gap worth noting: if you want to run from source, you will need to inspect the repository further to find the full build and launch commands. The bootstrap command suggests a one-time setup, likely installing dependencies and building native modules, but the documentation does not elaborate. For most users, the packaged binaries are the intended route. ## Scheduling, heartbeats, and background operation GooeyPi includes a scheduling system that works with the active harness. You can create one-time or recurring schedules, pick a project or existing session, choose a model, reasoning level, timezone, and Fast mode. You can run a schedule immediately, pause or resume it, and inspect the result and run history. This is a concrete feature that goes beyond a simple chat wrapper. On macOS, you can opt in under Settings → General to keep scheduled work running from the menu bar after closing the window, and optionally launch GooeyPi in the background at login. The packaged app also accepts `--background` to start without opening its window. This is useful for unattended automation. For Prime Agent, GooeyPi can manage Prime heartbeats alongside its own schedules when Prime Agent exposes them. The phrase 'when Prime Agent exposes them' is a caveat: the feature depends on the harness's own capabilities, so it may not be available in all versions. ## Browser control and annotations: a shared context The in-app browser is isolated, with an address bar, navigation history, downloads, and an external-browser handoff. You can point at page elements and add notes, and GooeyPi attaches those annotations to your next message. This is a distinctive feature: it lets you annotate a web page and have those notes passed to the agent as context. Additionally, you can enable browser control so an agent can open and operate its own tabs. You can watch its cursor and work in the same tabs when you want to take over. This is a form of human-in-the-loop control that is more interactive than simply pasting a URL. The README does not specify how the agent receives the annotations, whether as text or as structured data, but the workflow is clear. This feature is likely most valuable for tasks that involve web research or UI testing, where the agent needs to see what you are seeing. ## Capabilities and MCP: a guarded boundary The Capabilities page aggregates packages, plugins, extensions, skills, prompts, and MCP servers. GooeyPi can manage local stdio MCP definitions for OMP and adapter-enabled Pi. Network MCP and authentication stay under the selected harness's direct control; those entries are visible but read-only in GooeyPi, with bounded definition-only removal available for cleanup. This is a deliberate limitation. The README states that MCP discovery has an independent 2,500-definition-per-settings-file limit and reports an explicit warning when that limit is exceeded, so unrelated capabilities cannot hide supported MCP rows. This guardrail is sensible but also a hint that GooeyPi is not a full MCP manager. If you rely heavily on network MCP servers or need to edit their configurations from the GUI, you will still need to use the harness's own tools. The read-only status is a clear boundary: GooeyPi shows you what is there but does not let you change what it does not own. ## Voice input, desktop pet, and other extras GooeyPi offers voice dictation with OpenAI, Groq, Deepgram, or a local `whisper.cpp` installation. Optional voice keys are encrypted with the operating system's secure storage, and local dictation needs no API key. There is also a realtime voice companion for searching, discussing work, and starting a task in the selected harness. An animated desktop pet reacts while an agent works and carries the realtime voice controls. The included pets are Orb and GooeyPi, and it can discover compatible Codex pets. These features are optional and off by default, which is a good choice because they add complexity. The voice integration is not core to the workspace, but it shows the project's breadth. The desktop pet is a niche addition that may appeal to some users but could be seen as gimmicky. The README does not detail how the realtime voice companion works under the hood, so its reliability is unverified. ## Maintenance, licensing, and the trade-offs The project is MIT-licensed, which means you can use, modify, and distribute it freely, including in commercial products, as long as you preserve the license notice. The repository shows recent releases on a near-daily basis: v1.1.15 on 2026-08-23, v1.1.14 on 2026-08-22, and v1.1.13 on 2026-08-20. This suggests active maintenance, but the README does not include a changelog or upgrade guide. Upgrading likely means downloading the new release and installing over the old one, but the documentation does not specify whether settings and sessions migrate automatically. Given that each harness owns its sessions, GooeyPi's own settings, such as schedules and custom paths, may be stored in a local config that should persist across updates, but that is not confirmed. The main trade-off is that GooeyPi adds a layer of abstraction over tools that are already complex. If you are comfortable with the CLIs and do not need scheduling or browser control, the extra GUI may be unnecessary. But for users juggling multiple harnesses, the unified interface and background scheduling are concrete benefits. ## Conclusion Adopt GooeyPi if you already use Pi, OMP, or Prime Agent and want a unified GUI for streaming, scheduling, and browser control without moving credentials out of each harness. Skip it if you prefer the native CLI or need deep MCP network management, since those remain read-only. Before adopting, verify that your Node.js version meets the pinned 24.15.0 requirement if running from source, and confirm that the harnesses you use are detected in Settings → Harness, because the app only shows ready harnesses and custom paths must be set manually otherwise. ## FAQ ### What is GooeyPi? A desktop workspace that puts one interface in front of three separate coding agent harnesses on macOS, Linux and Windows. It leaves each harness in charge of its own models, logins and saved sessions, so switching between them changes the workspace without stopping work already running in another one. ### Which coding agent harnesses does GooeyPi support? Three. Pi is the base coding-agent harness in the Pi family, OMP adds its own plugins, commands and approval controls, and Prime Agent adds provider integrations and native heartbeats. You need only one to get started. ### How does GooeyPi find an installed harness? It checks common install locations automatically and the top-left switcher lists the harnesses that come up ready. If one is missing or lives elsewhere, the Settings page for harnesses can refresh detection or let you set a custom executable path. ### How do I run GooeyPi from source? You need Node.js 24.15.0 or newer and npm 12.0.2 or newer, with the Node version pinned in a version file. Run nvm install and nvm use, then npm run toolchain:bootstrap. The release preflight checks the toolchain before anything else, so a mismatched version fails immediately. ## Sources - [Official documentation](https://github.com/am-will/gooey-pi) - [Official README](https://github.com/am-will/gooey-pi#readme) - [Project repository](https://github.com/am-will/gooey-pi) - [Release notes](https://github.com/am-will/gooey-pi/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/am-will-gooey-pi