A detailed breakdown of what Codesu can do for you.
- Three agent kinds: Claude (launches the Claude Code CLI), Shell (a plain login shell), and Command (any custom program you specify)
- Workspaces: Organize agents by project folder, each with an accent color
- Agent States: Automatically derived — 🔵 working, 🔴 blocked, 🟢 done, ⚪ idle, ⚫ exited
- Audio Feedback: Optional chime when an agent finishes or gets blocked
- Keyboard Navigation:
Cmd+1…9to switch between agent tabs instantly - Global Shortcuts: Quick access to agent creation and every view (rebindable)
- Persistent Sessions: Claude agents resume exactly where they left off (via
claude --resume) - Reopen last closed agent:
Cmd+Shift+Zbrings back the agent you just closed
- xterm.js v6.0: Industry-standard terminal emulation
- WebGL Rendering: Smooth, responsive output even with heavy workloads
- Resize Support: Terminal panes adjust dynamically
- Keyboard Support: Full keyboard input including special keys and modifiers
- Output Coalescing: Optimized rendering for high-throughput sessions
- Custom Working Directories: Each agent runs in its workspace's folder
- Login shell: Shell agents launch your default login shell (from
$SHELL) - Graceful shutdown: Agents get SIGTERM then SIGKILL after a grace period, so Claude flushes and persists its session on exit
- Interactive Prompts: Automatic detection of blocking prompts from Claude
The Code view (Cmd + E) is the "I don't want to leave the app" half of the workflow: read what the agent wrote, fix it by hand, and run it — without opening a separate IDE.
- File tree of the active workspace, loaded one level at a time (a huge
node_modulescosts nothing until you expand it) with a "show hidden" toggle - CodeMirror 6 editor with syntax highlighting for JS/TS/JSX/TSX, JSON, CSS, HTML,
Svelte, Vue, Markdown, Rust, Python, Java/Kotlin and YAML — plus line numbers, code
folding, bracket matching and
Cmd + Fsearch - Tabs per open file, remembered per workspace across restarts (paths only — every buffer is re-read from disk, so you never look at a stale copy)
- Cmd + S to save, with an unsaved-changes dot on the tab
- Agent-aware: a file an agent rewrote in the background is re-read when you focus its tab, and a save is refused (offering Reload or Overwrite) if the file moved on underneath an edit of yours
- Sandboxed: every read and write is resolved against the workspace root and refused if it lands outside it, symlinks included
- Changes panel —
git statusfor the workspace, grouped by path with A/M/D/R/U badges, the current branch, and ahead/behind counts - Diff viewer with real old/new line numbers, per-file and "all changes at once"
- Stage / unstage a file straight from the list
- Auto-refresh every few seconds while the view is open, so an agent's commits and edits show up on their own
- Jump to edit: open the file being reviewed in the editor in one click
- Script discovery —
package.jsonscripts (run with the package manager the lockfile implies),Makefiletargets, Cargo, Gradle, Go, Python and the repo's shell scripts, including sub-projects up to two levels deep - Run panel — a real shell per workspace, so scripts run with your environment,
accept input, and keep their scrollback; Stop sends
Ctrl-C - Ad-hoc commands: type anything into the run bar
- The panel is collapsible and resizable, and keeps running while you are in another view
- Four-Column Board:
- Backlog: New tasks not started
- In Progress: Currently being worked on
- Testing: Tasks in QA/review
- Done: Completed tasks (auto-archive)
- (Ideas are captured separately on the Notes page and can be forked into tasks.)
- Drag & Drop: Move tasks between columns
- Task Details: Attach files, add descriptions, link agents
- Spawn an agent from a task: the task's title, details, and file paths become the agent's opening prompt
- Dedicated Agents: Assign agents to specific tasks
- Multi-Agent Tasks: Coordinate multiple agents on complex tasks
- Task History: See which agents worked on which tasks
- Task Comments: Attach notes and Claude responses to tasks
- Attachments: Add files to tasks (code, docs, images)
- Image Preview: View images directly in the app
- File Linking: Reference files across multiple tasks
- Workspace Files: Organize by project
- Automatic Tracking: Every action is logged
- Daily Summaries: See what you accomplished each day
- Timeline View: Hourly breakdown of work activity
- Task Metrics: Track productivity by task type
- History page: Review finished agents and restore them
- Task archive: Browse completed/archived tasks
- Day-by-day journal: When you worked on or completed each agent and task
- 120-Day Retention: The activity journal is pruned to a rolling 120-day window
- Task Search: Quickly locate tasks by name or description
- Archive browsing: Completed tasks stay searchable in History
- Markdown Support: Write notes in markdown format
- Formatting: Headers, lists, code blocks, links
- Auto-Save: Changes saved automatically
- Search: Full-text search across all notes
- Capture Ideas: Quick idea entry from anywhere
- Fork to Task: Convert ideas into actionable tasks
- Batch Processing: Create multiple tasks from related ideas
- Idea History: View all past ideas and their status
Defaults (macOS). Every shortcut is rebindable in Settings → Shortcuts.
- Cmd + A: Agents
- Cmd + E: Code
- Cmd + Y: Tasks
- Cmd + N: Notes
- Cmd + R: Daily Report
- Cmd + H: History
- Cmd + S: Settings
- Cmd + Shift + T: Toggle Terminal
- Cmd + T: New Claude agent
- Cmd + 1…9: Switch to agent tab 1–9
- Cmd + Delete: Close current agent
- Cmd + Shift + Z: Reopen last closed agent
- Cmd + 1 / 2 / 3: Switch between that page's sub-views (board / list / archive)
Keys typed inside a terminal pane (e.g. Ctrl+C, Ctrl+L) are handled by the
shell or CLI running there, not by Codesu.
- Create worktrees on the fly:
<repo>/.worktrees/<branch>, so parallel agents work on isolated branches without stepping on each other - List & remove: Manage a repo's worktrees straight from the app
- Branch Isolation: Each worktree is its own directory on its own branch
- Git Commands: Run any git command through agent terminals as usual
- Review in-app: the Code view's Changes panel shows
git statusand diffs for the active workspace, with staging — see Code View
Everything is serialized to a single JSON file (state.json), saved
automatically as you work:
- Workspaces & agents (metadata, session ids, working directory)
- Tasks and their status, plus notes / ideas
- Activity journal (rolling 120 days)
- Layout & shortcuts (sidebar/pane sizes, rebindings, per-view state)
Terminals themselves are runtime-only — they respawn when a tab is opened.
- macOS:
~/Library/Application Support/codesu/state.json - Windows:
%APPDATA%\codesu\state.json - Linux:
~/.local/share/codesu/state.json
- Atomic writes (temp → fsync → rename) so a crash can't corrupt
state.json - Rotating backup kept as
state.json.bak; auto-recovery if the live file is unreadable - Survives app upgrades and reinstalls (data lives outside the app bundle)
- Automatic migration from the old
com.kmlcnclk.codesu/folder
See the README's Your data & backups for details.
- Dark theme built in
- Workspace accent colors assigned automatically as you create workspaces
- Resizable sidebar, workspace/agent split, and notes pane — sizes are persisted
- Mute toggle for state-change sounds
- Automatic saving — no manual save, no interval to configure
- Default projects: preset folders offered when creating workspaces
- Keyboard shortcuts: fully rebindable, scoped per view; reset to defaults anytime
Codesu is built with Tauri 2, so it's a small native desktop app rather than a bundled browser.
- Primary target: macOS — universal build (Apple Silicon + Intel), native
window with a custom title bar,
Cmd-key shortcuts, and "open in VS Code / IntelliJ". - Data directories follow each OS's convention (Application Support on macOS,
%APPDATA%on Windows, XDG data dir on Linux). - Windows / Linux: buildable from source via Tauri, but not currently packaged or tested as first-class distributables.
Note: agent attention uses an in-app chime + sidebar re-sort, not OS-level notifications or a system-tray icon.
- Real-time team collaboration (share agents/tasks)
- AI-powered task suggestions
- Automated workflow templates
- Integration with external tools (GitHub, Linear, Jira)
- Custom agent templates
- Advanced analytics & insights
- Web-based dashboard
- Mobile companion app
- Memory: modest baseline; grows with the number of open terminal sessions
- CPU: minimal when idle; scales with terminal output (coalesced into ~8ms / 64KB batches)
- Disk: small install (~20MB app) plus your
state.json - Network: Codesu itself makes no network calls — only the
claudeCLI and any git commands you run reach the network. Agent state is read from terminal output, never polled from an API.
Codesu is constantly evolving. Check back for new features and improvements!