![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/ban_kan_logo_readme.svg\"){kind=logo}
\n\n \n
\nHandle 10+ AI coding agents without losing control.\n
\n\n\nDesign → Publish → Run → Inspect\n
\n\n\n The control center for managing many AI coding agents in one simple UI.\n
\n\n\nBring order to parallel AI development without leaving your local workflow.\n
\n\n\n ![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/workflow/taskflow_animated.gif\")
\n
\n ![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/bankan_screenshot.png\")
\n
\n CI\n ·\n GitHub\n ·\n Issues\n
\n\n\n⭐ If Ban Kan helps you ship faster, please consider starring the repo.\n
\n\n---\n\n## Installation\n\n### Run instantly\n\n```bash\nnpx @stilero/bankan\n```\n\n### Install globally\n\n```bash\nnpm install -g @stilero/bankan\nbankan\n```\n\n### Run from source\n\n```bash\ngit clone https://github.com/stilero/bankan.git\ncd bankan\n\nnpm run install:all\nnpm run setup\nnpm run dev\n```\n\nBan Kan starts a local server, opens your browser automatically, and serves the dashboard from the same process.\n\nContributor workflow, TDD expectations, pull request testing guidance, and verification commands live in [CONTRIBUTING.md](./CONTRIBUTING.md).\n\n---\n\n## Requirements\n\n- [Node.js](https://nodejs.org/) >= 22\n- [git](https://git-scm.com/)\n- At least one AI CLI tool:\n - [Claude Code](https://docs.anthropic.com/en/docs/claude-code) (`npm install -g @anthropic-ai/claude-code`)\n - [Codex CLI](https://github.com/openai/codex) (`npm install -g @openai/codex`)\n- [GitHub CLI](https://cli.github.com/) (`gh`) — required only for automatic pull request creation\n\nNative build tools may be needed if `node-pty` or `better-sqlite3` has to compile during install.\n\nmacOS: `xcode-select --install`\nLinux: `sudo apt-get install build-essential`\nWindows: Install [Visual Studio Build Tools](https://visualstudio.microsoft.com/visual-cpp-build-tools/) with the \"Desktop development with C++\" workload, or run `npm install -g windows-build-tools` from an elevated PowerShell\n\n---\n\n### Big heads up\n\nBan Kan is in early development.\n\nThe core workflow works, but bugs and rough edges are expected.\nFeedback is extremely valuable at this stage.\n\n---\n\n## Quick Start\n\n1. Launch Ban Kan\n\n```bash\nbankan\n```\n\n2. Complete the setup wizard\n\n3. Configure agent CLIs in `Settings -> Implementation` and `Settings -> Review` as needed\n\n4. Add one or more repositories in `Settings -> General -> Repositories`\n\n5. Review or customize the published workflow under `Workflows`\n\n6. Create a task and select a published workflow version\n\n7. Use the live task graph for interviews, approvals, artifacts, budgets, and run controls\n\n8. Optionally create a pull request\n\n---\n\n## What Is Ban Kan\n\nBan Kan is a **local control center for AI coding agents** that work across real repositories.\n\nInstead of one long AI chat trying to do everything, every new task snapshots an immutable, published workflow version. Phase nodes move its card through the Kanban board:\n\nBacklog → Planning → Implementation → Review → Done\n\nEach workflow can combine interviews, configurable agents, approvals, conditions, bounded loops, forks, joins, phases, actions, and terminal outcomes. Drafts may be incomplete, while publication requires deterministic routes, bounded cycles, executable-node policies, and provider-compatible model and permission settings.\n\nBan Kan ships four published workflows: **Standard Development** (the default), **Simple Linear**, **Security Focused**, and **Legacy Pipeline**. The workflow editor supports version history and secret-free JSON import/export. Existing nonterminal tasks loaded from `tasks.json` remain on the legacy executor; newly created tasks use the graph engine.\n\nWorkflow definitions and execution state are stored transactionally in `.data/workflows.sqlite`. Large transcripts and workspace contents remain filesystem-backed. A task snapshot never changes when its source workflow draft or later published versions change.\n\nBan Kan combines:\n\n- structured workflows\n- parallel agent execution\n- human approvals\n- local repository access\n- optional pull request automation\n\nAll in one dashboard.\n\n---\n\n## Why Ban Kan Exists\n\nWhen developers levels up running multiple AI coding agents they often end up juggling multiple terminals:\n\n- Agent 1 planning a feature\n- Agent 2 implementing code\n- Agent 3 reviewing changes\n- Agent 4 generating tests\n\nKeeping track of everything quickly becomes overwhelming.\n\nIn practice most developers struggle to manage more than 3–4 agents at once.\n\n\n ![\"Four](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/before_claude_windows.png\") \n \n Before Ban Kan\n \n Managing multiple agents means juggling separate terminal windows and fragmented context.\n | \n \n ![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/bankan_multi_tasks.png\") \n \n With Ban Kan\n \n Tasks, agent stages, approvals, and live output stay visible in one shared dashboard.\n | \n
\n ![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/workflow/add_task.png\")
\n
\n ![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/workflow/planning_stage_started.png\")
\n
\n ![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/workflow/planning_approve_plan.png\")
\n
\n ![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/workflow/implementing_task.png\")
\n
\n ![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/workflow/review_stage.png\")
\n
\n ![\"Ban](\"https://raw.githubusercontent.com/stilero/bankan/HEAD/docs/images/workflow/done_stage.png\")
\n