Skip to content
Tekai

Vibe Kanban — Kanban Interface for Administering AI Coding Agents

BloopAI (organizational) April 11, 2026 product-announcement medium credibility
View source

Referenced in catalog

Vibe Kanban Claude Code OpenCode OpenHands Beads (bd)

Vibe Kanban — Kanban Interface for Administering AI Coding Agents

Source: GitHub — BloopAI/vibe-kanban | Author: BloopAI (organizational) | Stars: 24.8k Category: product-announcement | Credibility: medium

Executive Summary

  • Vibe Kanban is an open-source (Apache 2.0) local-first web app that wraps AI coding agents with a kanban planning board, per-task git worktree workspaces, inline diff reviewing, and an embedded browser for live previews. The core pitch: “engineers spend most of their time planning and reviewing coding agents — describe the work, review the diff, ship it.”
  • The project launched recently and has accumulated 24.8k GitHub stars, 2.5k forks, and 358 open issues, signalling rapid adoption and an active but unstable development phase. The stack is Rust backend + React frontend with a monorepo structure, launched via npx vibe-kanban.
  • Vibe Kanban occupies a distinct niche: it is not an AI coding agent itself but an orchestration shell that sits above agents like Claude Code, Codex, Gemini CLI, GitHub Copilot, Cursor, Amp, and OpenCode. This framing is novel — positioning the developer’s role as planner/reviewer rather than coder.

Critical Analysis

Claim: “Get 10X more out of Claude Code, Codex or any coding agent”

  • Evidence quality: marketing claim (no benchmark, no controlled study)
  • Assessment: The “10X” claim is a marketing aspiration with no backing data. The underlying logic — that developers bottleneck on planning and review, not code generation — is plausible and aligns with emerging industry observations. However, the actual productivity gain depends heavily on task type, team size, and agent quality. There is no independent study comparing developer throughput with vs. without Vibe Kanban.
  • Counter-argument: Coordinating multiple parallel agents introduces its own overhead: workspace setup, credential management across agents, monitoring concurrent diffs, and resolving merge conflicts from parallel worktrees. For teams new to agentic workflows, the cognitive load of managing the orchestration layer may offset gains from parallelism.
  • References:

Claim: “Supports 10+ coding agents including Claude Code, Codex, Gemini CLI, GitHub Copilot, Cursor”

  • Evidence quality: feature claim (stated in README; integration depth varies per agent)
  • Assessment: The documented agent list (Claude Code, Codex, Gemini CLI, GitHub Copilot, Amp, Cursor, OpenCode, Droid, CCR, Qwen Code) is plausible given each agent’s CLI/subprocess interface. Vibe Kanban abstracts agents as subprocess commands rather than API integrations, which makes broad multi-agent support achievable without per-agent SDKs.
  • Counter-argument: The abstraction layer means Vibe Kanban can only control what agents expose through their CLI. Advanced features — like Claude Code’s CLAUDE.md hooks or Codex’s async delegation model — may not surface in the UI. Open issues mention specific failures: non-existent claude-code@2.1.62 package installation attempts and GitHub Copilot CLI slash command gaps confirm integration quality is uneven across agents.
  • References:

Architecture: Git Worktrees as Workspace Isolation

  • Assessment: Using git worktrees to give each agent a dedicated branch and working directory is architecturally sound. It enables genuine parallel execution without file-system conflicts and maps naturally to the “one task per branch” PR model. The implementation is consistent with how sophisticated agentic frameworks (e.g., OpenHands, multi-agent pipelines) handle isolation.
  • Known friction: “Workspaces disappear from kanban cards when local workspace is cleaned up” is a documented bug, suggesting the workspace state model and git worktree lifecycle are not yet fully consistent. The DISABLE_WORKTREE_CLEANUP env var for debugging is a signal that worktree cleanup is error-prone. Race conditions in diff rendering (long agent responses rendered twice) suggest concurrency handling is still being hardened.
  • References:

Claim: “Self-hosting with Docker, SSH, and tunnel support”

  • Evidence quality: feature claim (documented in README)
  • Assessment: The configuration options (VK_TUNNEL for relay mode, SSH-based remote access, VK_ALLOWED_ORIGINS for reverse proxy) are genuine infrastructure features enabling remote deployment. Docker support and npx-based single-command startup lower friction meaningfully. The relay tunnel mode addresses the common pain point of accessing a locally-running orchestrator from another machine.
  • Counter-argument: No independent production deployments or case studies were found. No RBAC, no audit logging, and no multi-user access controls are documented. Microsoft Entra ID OAuth support is listed only as a feature request, confirming enterprise auth is not yet implemented.
  • References:

Agent SDK Integration Concern: Settings Override

  • Assessment: A documented issue reveals “project-level .claude/settings.json hooks are overridden by SDK Initialize message.” This is a significant integration problem for Claude Code users who rely on pre-commit hooks, tool permissions, or PostToolUse hooks as guardrails. Vibe Kanban’s SDK initialization silently disabling configured hooks could result in agents executing without intended constraints — for example, bypassing hooks that block committing secrets or enforce linting.
  • Implication: Teams running Claude Code through Vibe Kanban may lose their hook-based security guardrails without realizing it. This is a trust boundary issue for production use.
  • References:

Where Vibe Kanban Fits in the Ecosystem

Vibe Kanban occupies a distinctive position at the intersection of three categories:

  1. AI coding agent shells (Claude Code, Codex, OpenCode) — Vibe Kanban wraps these, it does not replace them
  2. AI-aware issue trackers (Beads, Linear, GitHub Issues) — Vibe Kanban provides lightweight kanban planning but focuses on execution management rather than deep dependency tracking
  3. AI-aware code review tools (Graphite, GitHub PRs) — Vibe Kanban provides inline diff review within the same interface, removing context switching

The closest category peer is OpenHands, which also provides isolated workspaces for autonomous agents. The key difference: OpenHands builds its own sandboxed Docker runtime; Vibe Kanban orchestrates existing installed agents via subprocess, giving a lighter footprint but less isolation and less control over agent behavior.

Credibility Assessment

  • Author background: BloopAI is the company behind Bloop, a code search and understanding tool. Vibe Kanban represents a pivot toward AI agent orchestration tooling. No major venture funding announcement was found at time of review.
  • Publication bias: This analysis is based on the project’s GitHub repository, README, and issue tracker — not independent reviews or production case studies. The 24.8k stars reflect rapid community traction but do not indicate production adoption depth.
  • Verdict: medium — The project demonstrates real architectural thinking (worktree isolation, subprocess-based multi-agent abstraction, integrated diff review) and genuine community momentum (24.8k stars, 2.5k forks), but 358 open issues, documented integration gaps, and absence of enterprise features place it firmly in early-adopter territory. The settings override bug is a notable concern for teams relying on Claude Code hooks as security guardrails.