Vibeyard is an MIT-licensed open-source desktop IDE (Electron) built around agent CLIs rather than an editor: it wraps Claude Code, OpenAI Codex CLI, and Gemini CLI in managed sessions, adds a kanban board that spawns and closes those sessions, and tracks cost, tokens, and context per session in a live inspector.
Vibeyard’s bet is that the cockpit, not the editor, is the product: the CLIs keep doing the work, and the surface owns everything above them, sessions, tasks, costs, and sharing.
What it is #
An Electron and TypeScript desktop app where each agent session runs in its own PTY, with a drag-and-drop project dashboard (AI Readiness, Kanban, Team, Sessions, Provider Tools, live GitHub PRs and Issues widgets), a swarm-mode grid of every session at once, and a keyboard-first shortcut layer. The kanban loop is the differentiator: a card spawns or resumes a CLI session in one click, and tasks auto-move to Done when their session completes, so the board tracks what the agents actually finished. Multiple Claude profiles run side by side (work and personal licenses), each session backed by its own isolated config directory so credentials, settings, and history never mix. Cost, token, and context-window tracking run per session, the inspector (Cmd+Shift+I) adds a timeline, cost breakdown, and tool-usage stats, and smart alerts flag missing tools, context bloat, and session health issues. Sessions share over encrypted peer-to-peer WebRTC with PIN authentication, read-only or read-write, no server in between, and an embedded browser tab can feed an element’s selector, text, and URL to the agent as edit context. It is an independent one-maintainer project by elirantutia, distributed through signed and notarized builds, an npm one-liner, and Linux packages, and the README states it is not affiliated with or endorsed by Anthropic.
Status #
Quiet but shipped: 1,388 stars and 173 forks as of 2026-10-10, created 2026-03-19, released v0.3.4 (2026-07-22) through v0.3.8 (2026-09-03), and neither the repository nor the release train has moved since 2026-09-03, a five-week window as of this note’s date. The only public launch trace is a self-submitted Show HN thread from 2026-04-23 that drew 2 points and zero comments, and no third-party coverage surfaced in this run’s search. MIT throughout.
Strengths #
- The kanban-to-session loop is unique in this category: the board spawns the work and closes the card off the session’s own completion, so task state is derived from agent behavior rather than maintained by hand.
- Multiple isolated Claude profiles make running a work and a personal license side by side safe, which no other member documents.
- WebRTC session sharing with PIN auth is the only live share path in the category that needs no server.
- Cost, token, and context tracking per session plus the inspector put spend and window pressure next to the terminal.
- Signed and notarized desktop builds plus an npm one-liner keep install friction low.
Cautions #
- The five-week quiet window on both commits and releases is the status question this note cannot answer: one maintainer, no community, no stated roadmap for the pause.
- The community footprint is nearly empty, so nothing independent has exercised the feature claims (the P2P share path and profile isolation especially).
- It wraps three CLIs and inherits whatever they change; the README itself says more providers are “coming soon”, unshipped as of 2026-10-10.
- MCP and AGENTS.md support are not documented anywhere, so rules and tool servers stay managed inside the CLIs.
- No cloud, scheduled, or mobile story: sessions are local, and sharing ends at the WebRTC session.
Pricing #
Free and open source under MIT; no paid tier exists.
Compared to #
- OpenChamber: the incumbent session cockpit, with multi-model fusion and cron that Vibeyard lacks; Vibeyard answers back with the task board and the share path.
- Cate: the canvas-shaped cockpit; Cate organizes worktrees in space, Vibeyard organizes sessions in a board.
- Whiteboard: the review surface where finished agents draw their work; Vibeyard supervises the run itself instead.
Bottom line #
Recommended for engineers running Claude Code, Codex CLI, or Gemini CLI who want a task board and a cost view above their terminals, and who accept a single-maintainer app that has been quiet since September 2026. Not for anyone who needs a currently shipping cockpit (OpenChamber’s train is moving), MCP or rules management in the surface, or any cloud and mobile reach.
Changes #
- 2026-10-10 - Created from the category entrant scan, admitted on its differentiated kanban-to-session and share mechanisms with the quiet-window caution recorded.
See also #
- OpenChamber - the session cockpit incumbent this note joins as a challenger
- Cate - the canvas neighbor solving the same supervision problem spatially
- Whiteboard - the review-side canvas that complements the run-side cockpit
- Surface Feature Matrix - the category comparison this note joins
References #
https://api.github.com/repos/elirantutia/vibeyard - 1,388 stars, 173 forks, MIT license, created 2026-03-19, pushed 2026-09-03, as of 2026-10-10
https://raw.githubusercontent.com/elirantutia/vibeyard/main/README.md - the feature set (kanban-to-session loop, swarm grid, isolated Claude profiles, WebRTC sharing, inspector, embedded browser), install paths, and the not-affiliated notice
https://hn.algolia.com/api/v1/search?query=vibeyard&tags=story - the search that found exactly one relevant thread