Skip to main content

Kagura

Kagura is a local devflow app for a single developer's workstation. It pulls issues from your trackers — GitHub, Azure DevOps, Beads, or plain Markdown files — triages each one into a small set of parallelisable sub-tasks using the claude CLI, then runs those tasks as real, attachable claude sessions inside isolated git worktrees.

When every task lands, Kagura merges the work-item branch up and opens a pull request via gh.

Single-user, local-only, no auth. Built for the workflow of one developer juggling many concurrent agents.

What you can do with it​

  • Aggregate issues from multiple sources in one UI without context- switching between trackers.
  • Triage with Claude — turn a vague issue into 1–3 ordered tasks small enough for an agent to land in one sitting.
  • Run agents in parallel — each task gets its own git worktree and its own claude PTY session, capped by a configurable concurrency limit.
  • Attach, type, and steer — every PTY streams to an in-browser xterm.js tab over SignalR, so you can drop in to redirect an agent or unblock it.
  • Carry to PR automatically with Ralph Loop — one click that starts, merges, and PRs each task in order, with retries.

How it fits together​

┌──────────────────────────┐         ┌──────────────────────────────────┐
│  React + xterm.js (5173) │ <─SR──> │  ASP.NET Core API (5253)         │
│  Sources / WorkItems UI  │ <─REST─>│  + SignalR /hubs/agent           │
└──────────────────────────┘         │  + EF Core (SQLite)              │
                                     │  + claude CLI (triage)           │
                                     │  + Porta.Pty (claude PTYs)       │
                                     │  + git CLI (worktrees, PRs)      │
                                     └──────────────────────────────────┘
                                                  │
                                                  ▼
                                       ~/.devflow/kagura.db
                                       ~/.devflow/keys/
                                       ~/.devflow/worktrees/<wi>/<task>/
                                       ~/.devflow/transcripts/

The whole stack is orchestrated by .NET Aspire — one dotnet run starts the API, the Vite frontend, and a dashboard with logs, traces, and health for both.

Where to go next​

  • Setup — install prerequisites and boot Kagura on a fresh machine.
  • Usage — the four flows you'll touch most often: filtering, triage, the Closed status, and Ralph Loop.
  • GitHub repo — full source, the REST API reference, and the on-disk layout under ~/.devflow/.