🎨 Best DeepSeek Harness Design Plugin. The open-source Claude Design alternative. 🖥️ Local-first desktop app. 🖼️ Your coding agent becomes the design engine: prototypes, landing pages, dashboards, slides, images & video — real files, HTML/PDF/PPTX/MP4 export. 🤖 Claude Code / Codex / Cursor / DeepSeek Harness / OpenCode & 20+ CLIs via BYOK.
Desktop kanban board for agent CLIs
Kangentic is a desktop kanban app for agent-led work. You create a task card, drag it into a column, and the app starts the chosen agent with the right worktree, permissions, and context. It tracks sessions, handoffs, output, and cost across multiple agent CLIs.

Builders who want to run agent CLIs from a kanban board and keep work, context, and worktrees organized.
You can start, pause, resume, and hand off agent work from a single board instead of managing each CLI by hand.
What it does
Column-based workflows
Set up pipelines like Plan, Execute, and Review with per-column prompts, commands, permissions, and exit actions.
Live session tracking
See each agent’s status, output, progress, and notifications directly on the board.
Worktree isolation and review
Run each session in its own git worktree and open diffs and commit graphs from the board.
Session persistence and handoff
Save session state incrementally, resume after crashes, and move context between supported agents and columns.
Backlog and import
Bring in tasks from GitHub Issues, GitHub Projects, Azure DevOps, Asana, and other boards.
Command terminal and memory search
Open an ad hoc agent session, search past sessions, and find tasks and conversations with keyword or semantic search.
How to get it
- 1Run
npx kangentic
README
Kangentic
Drag a card. An agent starts.
A Kanban board for AI coding agents. Spawn, suspend, and resume sessions across fourteen agent CLIs from one board, with your own backlog. Local, free, open source. One board shows every agent's status, output, and progress: respond when needed, and let them work autonomously the rest of the time.
If Kangentic saves you time, hit ⭐ at the top of this page so others can find it.
Features
- Customizable workflows - build pipelines like Plan, Execute, Review, with permission modes, auto-commands, plan-exit targets, entry prompts, and exit scripts or PRs set per column.
- Real-time status - see which agents are thinking or idle right on the card, via native hooks where available and PTY fallbacks where not, with desktop notifications when one needs you.
- Agent Monitor - one overlay watches every project on your machine: live and recently finished sessions as cards, a table, or a dense list, each with a peek at the agent's latest output.
- Usage & cost analytics - tokens, cost, and burn rate by project, agent, model, and effort, over any time range, down to a per-project ledger with cost share and dollars per million tokens.
- Git worktrees & review - each agent runs in its own worktree, so parallel work never collides. The built-in Changes panel opens a split or inline diff with file tree and commit graph, one click from the card.
- Session persistence - session data is written incrementally, so even a hard crash loses nothing. On relaunch running agents auto-resume with full context, and sessions you paused stay paused.
- Handoff context - move a card from a Claude plan column to a Codex execute column and the next agent starts with the full history. Both directions for Claude, Codex, Gemini, Antigravity, Qwen, Kimi, Grok, and OpenCode.
- Model & effort routing - Opus at xhigh for Planning, Sonnet for Executing, another agent for review. Save ladders as named Board Profiles; Kangentic applies them live as cards cross columns.
- Project & global settings - every project carries its own agent, model, effort, permission mode, base branch, and worktree defaults, separate from machine-wide ones, in a searchable settings panel.
- Backlog, labels & priorities - stage work before it hits the board, tag it with custom labels and a fully-customizable priority scale, and batch-promote in one move. The tags keep working as board filters afterward.
- Agent-to-board tools - every session gets MCP tools to create tasks, move cards, add columns, search prior sessions, and even message another task's running agent, so agents self-organize.
- Quick Find & memory - Ctrl+Shift+F / Cmd+Shift+F searches tasks, backlog, session events, projects, and every past agent conversation, by keyword or on-device semantic memory. No API key.
- Terminal & activity log - a real xterm.js terminal per session with WebGL rendering, scrollback that survives restarts, and clipboard image paste, plus an Activity tab that distills output into a structured event feed.
- Command Terminal - Ctrl+Shift+P / Cmd+Shift+P opens an ephemeral agent session over any board, no task card needed. Run up to four tiled side by side, with a layout that persists across projects and restarts.
- Context Bar - a live strip under every terminal: model, cost, tokens, tool calls, elapsed time, context-window fraction, and Claude rate-limit meters. The profile, model, and effort pills double as pickers.
- Embedded browser - point a sandboxed Chromium pane at any URL inside the task dialog, annotate it, and send the frame to the agent. Agents can screenshot, inspect, and click the same pane through MCP.
- Voice dictation - hold your push-to-talk button, talk, release: on-device speech-to-text drops your words wherever you are focused, a text field or the agent's terminal, with a streaming preview and a refinement pass.
- Your CLIs, your machine - runs on your desktop (Windows, macOS, Linux, and WSL) with your data in a local database. No OAuth, no wrappers, no API proxies: Kangentic launches the native CLIs you already have, with your own logins and subscriptions.
How It Works
- Create a task - add a card with a title and prompt. Paste screenshots, choose a source branch, and toggle worktree isolation, all from the create dialog.
- Drag to run - drag the card to any active column. Kangentic creates a worktree, picks the permission mode, and spawns your chosen agent automatically. Columns ship preconfigured, To Do through Done; reshape the pipeline, agents, and permissions per column anytime.
- Watch it code - your agent starts writing immediately. Follow along in the live terminal: see diffs, test results, and tool calls as they happen. Drag between columns to steer, or drag to Done to pause and pick up later.
Supported Agents
Fourteen coding-agent CLIs, all first-class, on one Kanban board. Mix agents per column and hand off context between them:
- Claude Code (Anthropic)
- Codex CLI (OpenAI)
- Gemini CLI (Google)
- Antigravity CLI (Google)
- Qwen Code (Alibaba)
- Kimi Code (Moonshot AI)
- Grok Build (xAI)
- OpenCode (sst)
- Droid (Factory)
- Cursor CLI
- GitHub Copilot CLI
- Aider
- Oz CLI (Warp)
- Ollama (local models)
Supported Boards
Bring your own backlog. Pull tasks in from the tools your team already uses, including titles, descriptions, labels, and inline images. Already-imported items are detected automatically so re-syncing is safe:
| Board | Status |
|---|---|
| GitHub Issues | Supported |
| GitHub Projects | Supported |
| Azure DevOps | Supported |
| Asana | Supported |
| Jira | Coming soon |
| Linear | Coming soon |
| Trello | Coming soon |
| GitLab | Coming soon |
| Obsidian | Coming soon |
Prerequisites
- Node.js 20+ (for npx)
- Git 2.25+
- At least one supported agent CLI (see Supported Agents)
Setup
npx kangentic
One command to download, install, and launch. After the first run, auto-updates handle everything.
For more details, see the Installation & Setup guide.
Documentation
Get started at kangentic.com/getting-started.
Development
Building from source requires Node.js 22+ (the npx floor above is for end users running the launcher).
git clone https://github.com/Kangentic/kangentic.git
cd kangentic
npm install
npm start
See CONTRIBUTING.md for project structure, testing, and conventions.
Contributing
See CONTRIBUTING.md for guidelines. All contributors must sign a CLA before their first PR can be merged.
Support
- GitHub Discussions for questions and feature requests
- GitHub Issues for bug reports
- support@kangentic.com if you would rather not post publicly
Found a security issue? See SECURITY.md instead.
License
AGPL-3.0. If AGPL doesn't work for you, drop us a line at licensing@kangentic.com.
Built with
Files in the repo
- .claude
- .github
- .husky
- build
- docs
- packages
- scripts
- src
- tests
- .eslintrc.json
- .gitattributes
- .gitignore
- .mcp.json
- announcements.json
- CHANGELOG.md
- CLA.md
- CLAUDE.md
- CODE_OF_CONDUCT.md
- CONTRIBUTING.md
- electron-builder.yml
- eslint.config.mjs
- index.html
- kangentic.json
- LICENSE
- package-lock.json
- package.json
- playwright.config.ts
- README.md
- RELEASE_NOTES.md
- SECURITY.md
- tsconfig.json
- vite.config.mts
- vitest.config.ts
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More other
"Vibe-Trading: Your Personal Trading Agent"
OpenHuman is an open source personal AI for Mac, Windows and Linux — local-first memory, agent orchestration, and deep research.

Your Personal AI Assistant; easy to install, deploy on your own machine or on the cloud; supports multiple chat apps with easily extensible capabilities.
User-friendly AI Interface (Supports Ollama, OpenAI API, ...)
macOS video editor built for AI
