Skip to content

Kangentic Documentation

Kangentic is an agentic development environment (ADE): a cross-platform desktop app where a Kanban board runs your AI coding agents. It treats thirteen coding-agent CLIs as first-class citizens - Claude Code, Codex, Gemini, Cursor CLI, GitHub Copilot CLI, Grok Build, Antigravity CLI, OpenCode, Aider, Qwen Code, Kimi Code, Factory Droid, and Warp (Oz CLI) - plus Ollama for local LLMs. Instead of managing agents one at a time in separate terminals, you spawn, monitor, suspend, and resume them all from one board, each task isolated in its own git worktree.

AI agents can handle complex tasks autonomously, and with git worktrees you can run many in parallel. But as agents get more capable, the bottleneck shifts to the human - managing context across multiple agents, knowing what each one is doing, and catching when they need attention. Kangentic puts every running session on one Kanban board and flags the ones that need you, so you stay in the loop without slowing agents down.

  1. Create a task - Add a card to your board describing the work (bug fix, feature, refactor)
  2. Assign and run - Drag the card to the Executing column to spawn the column’s agent
  3. Monitor progress - Watch the agent work in a real-time terminal, or let it run in the background
  4. Review and merge - When the agent finishes, review the diff in the built-in Changes panel and merge

Each agent session runs in its own git worktree with a dedicated terminal, so you can run many agents in parallel without interference.

Kangentic is deliberately unopinionated about three things. The agents you run, the work you feed it, and the process you run that work through are all yours.

Pick a default agent per project, override it per column, and hand work between agents with their context intact. Sessions survive crashes and restarts, and Kangentic reads each agent’s own activity signals so the board can tell you which ones are thinking, which are idle, and which are waiting on you.

Stage work before it reaches the board, with labels, priorities, and filtering, then promote a week’s worth to any column in one move. Import from GitHub Issues, GitHub Projects, Azure DevOps, and Asana; re-syncing is safe because already-imported items are detected automatically.

  • Backlog - staging, labels, priorities, bulk operations, and importing

The seven columns Kangentic seeds are a starting point, not the product. Rebuild them: each column carries its own agent, model, reasoning effort, permission mode, auto-command and its timing, spawn behavior, handoff, session target, and plan-exit target. Share the result with your team through kangentic.json, and keep personal deviations in kangentic.local.json.

The workflow runs both ways. Every spawned session holds the board’s own MCP tools, so an agent can search past conversations, message another task’s agent, and move its own card when the work is done.

FeatureDescription
Visual task managementA Kanban board where each card represents an agent session
One-click agent controlDrag cards to spawn, suspend, or resume agents
Isolated environmentsEach agent gets its own git worktree, preventing conflicts
Review built inA diff viewer with file tree, commit graph, and blame, one click from the card
Conversation memoryOn-device keyword and semantic search over every past agent conversation
Real-time monitoringWatch all agents work with integrated terminals, activity logs, and the machine-wide Agent Monitor
NotificationsDesktop alerts and idle badges when agents need attention
Multi-project supportManage multiple repositories from a single sidebar
Mobile companionCheck on agents, answer permission prompts, and review diffs from your phone
Local-firstRuns on your desktop with the agent CLIs and credentials you already have. Your boards, tasks, and transcripts stay in a local database, never on a Kangentic server. See the Privacy Policy

Kangentic is free and open source. A star helps other people find it.

Star on GitHub