early access · macOS first

Run parallel coding agents
in one desktop window.

hang4r wraps the Claude Code, Codex, and Cursor CLIs you already pay for. Parallel sessions in isolated git worktrees — streaming chat, diffs, terminal, review — with your hooks, skills, subagents, MCP, and subscription intact.

⌘⇧P 5H ● 1h 58m SEVEN_DAY ● 71h 59m COST $0.42 TOKENS 12.4k↓ 3.1k↑
fix flaky e2e ○ hang4r FilesDiffTerminal

Fix the flaky session-restore e2e test

Working on it — turn 1.

Worked 1 step

Subagent scanned the repo and found the fixed 500 ms wait in restore.spec.ts.

Edit e2e/restore.spec.ts

done · 12.4s · $0.0123

Changes 1 Create Branch & Commit Send follow-up… (@ file · / command) Send
claude⌥ worktreeopusacceptEdits 1.2k↓ 340↑ $0.012
polish diff viewer ○ hang4r FilesDiffTerminal

CHANGED FILES (1) Commit Merge PR

M src/components/DiffView.tsx +2 −1

@@ -212,3 +212,4 @@

- setTimeout(apply, 500)

+ requestAnimationFrame(apply)

+ // paint before Monaco re-measures

you use rAF here too? → sent to agent
Changes 1 Create Branch & Commit Send follow-up… (@ file · / command) Send
claude⌥ worktreesonnetplan 2.8k↓ 610↑ $0.019

inside the window

Everything in flight, at a glance.

Parallel sessions tiled side by side, diffs you can talk back to, and your account limits always in view — this is what a day of orchestrating agents actually looks like.

hang4r tiled workspace: two agent sessions streaming side by side with sidebar, usage gauges, and per-session status bars
Two agents mid-task, side by side — each tile is a full session with its own chat, files, diff, and terminal, while the sidebar keeps every project and rate limit in peripheral vision.
hang4r diff review panel showing changed files and an inline diff for an agent session
Diff review per session — changed files, hunks, and inline comments that go back to the agent as a follow-up.
hang4r new-agent dialog with backend, model, permission mode, and worktree isolation options
Dispatch an agent — pick Claude, Codex, or Cursor, choose model and permissions, isolate it in a worktree.

the loop

Dispatch. Glance. Review. Merge.

01

A tiled workspace for many agents

Sessions live side by side — up to four tiles per workspace, grouped by project in the sidebar with live status. Kick off an agent in one project while three others stream in parallel, and queue follow-ups while they work.

02

Every session in its own worktree

Isolation by default: each agent works on its own branch with per-turn checkpoint commits, so you can rewind any turn — or race the same task across Claude, Codex, and Cursor with /best-of-n and keep the winner.

03

Review that talks back

A changed-files panel and diff viewer where your inline comments are fed back to the agent as a structured follow-up. Annotate the diff, hit send, and the agent revises — then stage, commit, or open the PR.

04

Your CLI, not our API keys

hang4r drives the claude, codex, and cursor-agent binaries you already have as subprocesses. Hooks, skills, subagents, MCP servers, and subscription billing all keep working — we never touch ~/.claude.

mission control

Everything around the agents, built in.

⌘F everywhere

One find bar across conversation, editor, and terminal — same keys, same highlights.

Session import

Your existing Claude Code, Codex, and Cursor session history shows up on day one — continue any of it in hang4r.

settings.json

VS Code-style app + workspace settings with schema validation — your comments survive every save.

SSH remote

Run agents and terminals on a remote host over SSH tunnels; the UI stays local.

Usage gauges

Claude 5-hour and weekly limits, per-session tokens and cost — always visible, never a surprise.

Checkpoints & rewind

Per-turn commits on the session branch. Any turn can be rewound, nothing is lost.

Hooks timeline

Every hook that fires, in order, per session — see the machinery working.

Subagent inspector

The full subagent tree with each child's transcript, not a black box.

Queued messages

Type while the agent works — follow-ups queue and dispatch in order.

Notifications

Configurable per event, with action-required badges you can't miss on the dock and sidebar.

Terminal · editor · browser

Real PTY terminals with splits, a Monaco editor, and an embedded browser per session.

Processes panel

Every subprocess an agent spawns, live — inspect them, and kill the whole group cleanly.

how it compares

The agent-manager landscape, honestly.

Compiled July 2026 from public docs, changelogs, and repos. Corrections welcome — [email protected].

Feature hang4rmacOS · free EA CLI + tmux + nvimthe DIY stack · free ConductormacOS · free CursorIDE · $20–200/mo Claude Code / Codex appsfirst-party Vibe KanbanOSS · community claude-squadtmux TUI · AGPL
Runs on your existing CLI subscription ✓ it IS the CLI✗ own billing~ its own agent
Claude + Codex + Cursor in one app ✓ any CLI, by hand✓ +OpenCode~ multi-model, one vendor✗ single agent✓ 10+ agents
Git worktree isolation per session ~ manual / scripts
Diff comments sent back to the agent ✗ describe it in prose~ via PR @cursor
Built-in code editor + real terminal ✓ Monaco + PTY✓ nvim IS the editor~ terminal only✓ full IDE~ local sessions~ terminal + preview
Per-turn checkpoints & rewind ~ Claude's /rewind only~ Claude ✓ · Codex ✗—*
SSH to your own remote host ✓ ssh + tmux, natively✓ + cloud✓ + cloud VMs~ Claude ✓ · Codex cloud-only~ self-host
Rate-limit & per-session cost gauges ✓ all sessions, one view~ /status, one pane at a time—*~ dashboard~ usage ring / status—*
Imports your existing CLI session history ✓ it lives there—*~ own history—*
Hooks timeline & subagent inspector ✗ scrollback archaeology—*~ partial—*

shipped  ·  ~ partial or different shape  ·  not offered  ·  —* not documented publicly as of July 2026 (absent from docs, not proven absent).

Context worth knowing: Crystal was deprecated in Feb 2026 (succeeded by the paid Nimbalyst); Vibe Kanban's company shut down in April 2026 and the project is now community-maintained; Sculptor isolates agents in Docker containers instead of worktrees; opcode has shipped no releases since 2025. Cursor's agents bill through Cursor plans rather than your Claude/ChatGPT subscription. Anthropic and OpenAI both ship first-party session GUIs now — each drives only its own agent, which is exactly why a multi-backend, local-first window exists.

And full respect to the DIY column — raw CLIs in tmux splits with neovim is how a lot of this work gets done today, and it holds up better than most products here. hang4r exists for the day you're running five of those panes and can't tell, at a glance, which one is blocked waiting on you.

from the blog

Notes on agentic engineering.

why we're building it

Cursor's Agents Window proved the shape: the control plane for agents, not the text editor, is the primary surface. hang4r takes that shape and grounds it in the CLIs you already trust — open in approach, desktop-native, subscription-friendly.

MIT-clean

Learn from everything, copy nothing.

Local-first

Your repos, your machine, your keys — SSH when you want it.

Transparent

Usage gauges, hooks timeline, subagent inspector. See the machinery.

The hang4r doors open soon.

First builds go to the early-access list. No spam — a confirmation now, one email when your build is ready.