Mission control
for Claude Code
Monitor every session, launch every project, rescue every context — and share work with clients and partners. One dashboard, running entirely on your machine.
Live mission control
Run five Claude sessions. Babysit none.
Claude Helm watches your local Claude Code history and shows what every session is doing right now — and steps in at the two moments that actually cost you: a stalled session, and a full context window.
Every live session, named
The ambient line lists each running session — what it's working on, whether it's waiting on you, and how full its context window is. Click any of them to jump in.
"Waiting for you" pings
A session finishes its turn while you're in another window? Desktop notification plus a tray badge. Sessions never sit stalled again.
Context guard
When a session's context passes 85%, a card explains it in plain words and offers a one-click rescue: the full conversation memory is written to HANDOFF.md, then a fresh 200K-context session picks up exactly where you left off. Nothing is ever lost.
Quiet ambient awareness
Session-finished and budget notifications, plus a system-tray summary of today's time and cost — without opening the window.
Launch anything
From dashboard to running code in one click
Every project is launchable — into Claude, into a browser, or into a headless task. The dashboard is where work starts, not just where it's measured.
Open in Claude
A terminal already in the project, running claude — with per-launch model overrides. Windows Terminal, macOS Terminal/iTerm, and every common Linux terminal.
Launch the app inside the project
Anything with an index.html or a dev script gets a Launch button: static sites get a built-in local server, dev servers boot with their URL captured automatically. Opens in an in-app preview window with reload and browser pop-out.
Share a preview with a client
One click tunnels a running preview to a public https link with a QR code — your client opens your local site on their phone before anything deploys. The link dies when you stop it.
Quick tasks
Type a task into any project's card — "fix the mobile nav overlap" — and it runs headless claude -p in that project. A tasks strip tracks progress; you get a ping when it's done.
Resume or branch any session
Reopen an exact past conversation, or fork it into a new session and explore a different direction with the original untouched.
Agent maker
Build Claude Code subagents in-app — global or per-project, with tool pickers, model selection, and AI-drafted system prompts. They appear in Claude Code automatically.
Search everything
Any description finds the thing
One search across projects, conversations, and Claude's memory — with word-AND matching, so "roofing client site" matches even when those words never sit next to each other.
Projects pull right up
Search by a project's README, your notes, its tag or client, or a session title — the project appears at the top with Open and details one click away.
Suggested searches
The empty search tab recommends searches mined from your own usage: recurring session themes, recent projects, most-edited files, top tools. It guesses what you were about to type — because it watched you work.
Transcript viewer
Read any conversation as a clean chat — your messages, Claude's, tool calls — then Resume or Branch it from right there.
Analytics & money
Know what every project costs
A live engine tails your transcripts and computes time, cost, tokens, tools, files, and models per project — honestly labeled as estimates, with an optional Admin API key for real billed numbers.
Live project cards
Time, cost, activity sparkline, a live "active" dot, and git status on every project — plus a deep-dive detail view with per-day charts, token breakdowns, and tool usage.
Recap
What you did today or this week, across every project — with an optional AI-written narrative standup.
Budgets & spend
Weekly and monthly budgets with 80% and 100% alerts, spend by model, busiest-hours heatmap, side-by-side project compare, and CSV export. A dedicated MCP tab tracks every server, tool, and call across your projects.
Your real plan
Settings shows the actual Claude plan you're signed into — read locally from Claude Code's own session. Your login token is never read or transmitted.
Personalization
Eight gradient themes, one floating dashboard
Translucent glass panels float on accent washes — pick a theme and the entire UI recolors with its gradient, on light or dark. The burst logo stays Claude-orange in all of them.
Privacy
Local by design
Claude Helm reads ~/.claude on your machine. Transcripts never leave it. No telemetry, no account. API keys are encrypted with your OS keychain. The only network calls are ones you ask for: update checks, optional AI summaries, preview-share tunnels, and partner-project git sync.
Download
Get Claude Helm
Grab the latest release. The apps aren't code-signed yet, so each OS shows a one-time prompt.
| Platform | File | First run |
|---|---|---|
| Windows | Claude-Helm-Setup-*.exe | SmartScreen → More info → Run anyway. Updates itself silently after that. |
| macOS (Apple Silicon) | Claude-Helm-*-arm64.dmg | Drag to Applications, then run the unblock command below once. |
| macOS (Intel) | Claude-Helm-*-x64.dmg | Same as above. |
| Linux | Claude-Helm-*.AppImage | chmod +x then run. |
macOS says "damaged and can't be opened"? That's the quarantine flag on unsigned downloads, not corruption:
xattr -dr com.apple.quarantine "/Applications/Claude Helm.app"