Kangentic/kangentic

Drag a card. An agent starts.
An agentic development environment for AI coding agents.
Live demo |
Quickstart |
Documentation
## What it does
A coding agent can carry a task on its own, and git worktrees let several run at once. The limit then becomes you: which agent is waiting on a prompt, which one finished, and where each branch went. Kangentic puts every session on one board and marks the ones that need you.
### Run your agents from one board
One board, every agent. Each card shows its model, how much of the context window is used, and what the agent last said, with a mark on the ones that need you.
Kangentic Mobile. Pair your phone once, then check on agents, answer the question one stopped on, and review its diff away from your desk. The connection is end-to-end encrypted, through Kangentic's relay or one you host. Get it on the App Store or Google Play.
Fifteen agent CLIs. Plan with Claude Code and build with Codex, or use one CLI everywhere. Kangentic runs the CLIs you already have with the credentials you already configured. No API key, no proxy, no Kangentic account.
Every project in one place. Each repo gets its own board, and the Agent Monitor shows the agents running across all of them.
Review on the board. The Changes panel shows each task's diff and commit history. A linked pull request on GitHub or Azure DevOps shows its state and whether it is ready to merge.
Free and open source. AGPL-3.0, with every feature included and no paid tier.
### Shape the workflow
Columns. Rename, add, and rewire columns until the board matches how you ship. Each column carries its own agent, model, effort, and permission mode. The setup commits with the repo in kangentic.json, so teammates get the same board, while tasks and agent sessions stay on each person's machine.
Automations. When a task enters a column, when its agent finishes a turn, or when the task leaves, a column can send the agent a message or slash command, run a script, call a webhook, notify you, or move the task to the next column. A condition in plain words, checked by an agent, decides whether a step runs.
Workflows. Give the workflow agent a brief, review the linked tasks it plans, and press Start. Each task starts when the tasks it waits for reach Done, and one workflow can span several projects.
Agents that work the board. Kangentic's MCP server lets a running agent create tasks, move its own card, search past conversations, and message another task's agent, so a rule written in a prompt can move work along.
## Is it right for you?
It's a good fit if you:
run several coding agents at once and lose track of which one is waiting on you.
juggle tasks across several projects or clients and want every agent in one view.
want to mix Claude Code, Codex, and other CLIs on plans you already pay for.
have a multi-step workflow and want to customize its agents and automations per project.
want to answer your agents from your phone when you step away.
It's probably not for you if you:
prefer to work on one task at a time, start to finish.
work in a single project and rarely switch.
don't use agent CLIs, or prefer apps like Cursor or Claude Desktop.