ralph-ban
TUI kanban board for agents, built in Go with Bubble Tea. Five columns (Backlog, To Do, Doing, Review, Done), vim-style navigation, real-time sync across CLI/Claude Code hooks/TUI sessions via 2-second polling, SQLite backing. "A ralph loop sets a target and works until it's done. ralph-ban makes that target a kanban board."
Key Quotes
"A ralph loop sets a target and works until it's done. ralph-ban makes that target a kanban board."
Key Themes
The "ralph loop" concept is simple and effective: set a target, work until completion. The kanban board gives that loop visual structure and persistent state. The Claude Code integration (hooks + agents) means the board isn't just a display -- it's an active coordination surface where agents can claim and move cards.
Three usage modes: interactive TUI (manual management), batch mode (human approval for agent automation), and auto mode (unattended processing). The progression from human-in-the-loop to autonomous is well-designed.
Part of the same ecosystem as workgraph (which provides the more sophisticated task graph), weft (Cloudflare-hosted alternative), and Managing Agents via Kanban Boards (Geoffrey Litt's approach). ralph-ban is the simplest of these -- a kanban board and nothing more, which might be exactly the right level of complexity for most projects.
Critical Analysis
Sometimes the right tool is the one with the fewest features. ralph-ban doesn't try to be a task graph, an orchestration system, or a project management tool. It's a kanban board with agent integration, and that's enough. The 2-second polling for sync is pragmatic but will cause issues at scale -- SQLite can handle it, but the polling interval means cards can briefly exist in inconsistent states across sessions. For 1-3 agents on a personal project, this is ideal. For anything bigger, look at workgraph.
Sources: ralph-ban Last updated: 2026-05-14