Vibeyard
An IDE purpose-built for AI coding agents, making multi-session management, cost tracking, and task planning effortless.
Evidence: Local desktop app, no network services; SECURITY.md states threat model assumes trusted local user; Electron sandbox disabled but context isolation enabled; temp directory /tmp/vibeyard/; dependencies include electron-updater. Deductions: Sandbox disabled and temp directory use lack mitigation details; no user confirmation mechanisms found; data flow transparency limited; sensitive data handling (e.g., credentials) not detailed; dependency security lacks audit or lockfile; external effects (e.g., auto-update) not user-controlled; rollback not mentioned; source attribution only via author info, unverified.
Evidence: package.json scripts and dependencies consistent; CI workflow builds and tests on multiple OSes; test stubs handle Electron dependency. Deductions: Failure messages not documented; dependency availability lacks lockfile or mirror; self-consistency slight mismatch (README claims multi-platform, SECURITY.md mentions unsigned builds).
Evidence: README clearly targets AI coding agent users, supports multiple CLI tools; capability boundaries described in feature list; environment fit covers macOS, Linux, Windows, npm. Deductions: Trigger precision (e.g., shortcuts) not detailed; scenario examples limited; environment fit lacks specific version requirements (Node v24+ vs package.json >=18).
Evidence: README provides install instructions, feature list, contributing guide, license; LICENSE file MIT; SECURITY.md lists known limitations; package.json includes version and repo info. Deductions: Information architecture lacks directory or doc structure; naming stability not stated; examples and FAQ missing; versioning changelog absent; maintenance responsibility unclear (only author info).
Evidence: Output usability shown via feature list and demo GIFs; marginal value via unique features (P2P session sharing, cost tracking); cost-benefit not quantified. Deductions: Cost-benefit not quantified; output usability lacks real-world use cases.
Evidence: README claims partially traceable to code (e.g., feature list); cross-source corroboration limited (only README and SECURITY.md); facts and inferences not clearly separated. Deductions: Claims lack tests or independent verification; no evidence links provided.
- Sandbox disabled and temp directory usage may pose security risks; users should assess.
- No user confirmation mechanisms; sensitive operations may execute automatically.
- Dependencies not locked, supply chain risk.
- Auto-update mechanism lacks user control options.
- No rollback mechanism; update failures may impact usage.
What does this agent do, and when should you use it?
Vibeyard is a desktop IDE designed specifically for running AI coding agents like Claude Code, Codex CLI, and Gemini CLI. It provides a robust workspace with multi-session parallel management, real-time cost and context monitoring, a drag-and-drop project dashboard, a kanban task board, peer-to-peer session sharing, and session resume functionality. Built with Electron, it offers prebuilt installers for macOS, Linux, and Windows, as well as an npm global install option. Vibeyard itself is not an AI agent but a UI layer that wraps the agent CLIs, integrating with existing tools via PTY and WebRTC. Its MIT license and active release stream make it suitable for developers who want to manage multiple AI coding sessions more efficiently.
Built on Electron, Vibeyard launches and manages multiple PTY sessions, each running an AI CLI tool (Claude Code, Codex CLI, Gemini CLI). Users can create multiple sessions per project, run them in parallel, spin up new ones with Cmd+\, and use a grid view for swarm mode. The app collects and displays per-session cost, token usage, and context window occupancy in real time, plus a session inspector (Cmd+Shift+I) with timeline, cost breakdown, and tool usage stats. The project dashboard supports drag-and-drop customization with widgets like AI Readiness, Kanban, Team, Sessions, Provider Tools, and GitHub PRs/Issues. The kanban board allows dragging cards, launching or resuming sessions in one click, and auto-moves tasks to Done when their session completes. P2P session sharing uses WebRTC with PIN authentication and read-only/read-write modes. An embedded browser tab lets users preview localhost and send AI editing instructions with element selectors. Session resume functionality restores previous states after an app restart.
- A developer runs multiple Claude Code sessions simultaneously on different feature branches, monitoring progress via split panes and swarm mode.
- A team plans coding tasks on a kanban board, each task spawning an AI agent session and auto-updating status when the agent finishes.
- A developer tracks AI coding tool expenses and token consumption per session to stay within budget.
- Team members collaborate in real time via P2P session sharing, allowing remote peers to view or assist the active terminal session without screen sharing.
- A project maintainer uses the AI Readiness score and smart alerts to check for missing tools or context bloat and applies one-click fixes.
- A developer opens a local dev server in a session tab, clicks DOM elements, and instructs the AI agent to edit based on the exact selector.
What are this agent's strengths and limitations?
- Designed for multi-agent sessions, with parallel execution and swarm grid view, boosting productivity.
- Integrated cost and context tracking helps control token usage and expenses.
- Customizable dashboard and kanban board deeply integrate project management with agent execution.
- Peer-to-peer session sharing enables real-time collaboration without extra infrastructure.
- Multiple Claude profile isolation allows simultaneous work and personal account usage.
- Requires pre-installing and configuring specific CLIs (Claude Code, Codex, Gemini CLI), relying on external authentication.
- Electron-based desktop app with per-OS downloads; a large installer footprint.
- Platform-specific to AI CLI providers; switching providers requires reconfiguration.
- Steeper learning curve for beginners due to multi-session management and keyboard shortcuts.
How do you install or deploy this agent?
### PrerequisitesInstall and authenticate at least one supported CLI: Claude Code, OpenAI Codex CLI, or Gemini CLI.
### Installation Steps- macOS: Download the latest .dmg from GitHub Releases, drag to Applications, and launch (signed and notarized).
- Linux: Download the .deb (Debian/Ubuntu) or .AppImage (universal); run sudo dpkg -i vibeyard_*.deb or chmod +x Vibeyard-*.AppImage && ./Vibeyard-*.AppImage.
- Windows: Download the Setup .exe or portable .exe; run the installer or execute the portable directly.
- npm: Install globally with npm i -g vibeyard, then run vibeyard; it auto-downloads and launches on first run.
- From source: git clone https://github.com/elirantutia/vibeyard.git && cd vibeyard && npm install && npm start, requiring Node.js 24+.
How do you use this agent?
After installing and launching, select or create a project folder. In the dashboard, add widgets and create kanban tasks. Press Cmd+\ or click the new session button to start an agent session, choosing the CLI (Claude Code, Codex CLI, or Gemini CLI) and profile (supports multiple Claude configs). Monitor output, cost, and token usage in the session tab. Use the session inspector (Cmd+Shift+I) for detailed telemetry. To share a session, use the P2P share feature and generate a PIN. Kanban cards update automatically when sessions complete.