Run 5 or 25 Claude Code sessions and see them all in one place: what each one is doing, which ones need you,
what they changed and what they cost. Only the things that need you interrupt you.
Running a handful of Claude Code sessions at once means a wall of terminals. One of them is waiting on a permission prompt. Another finished ten minutes ago. Two are editing the same file. You find out by tabbing through every window.
tend turns that wall into one page. It doesn't wrap or replace Claude Code: every session is the real Claude Code TUI running in tmux. tend listens to Claude Code's own hooks and status line, and puts what matters on top.
| For you lists the sessions waiting on a permission, a question or a plan, then problems (stuck, crashed, overlapping). Everything else stays quiet. | Permission, question and plan cards go through Claude Code's own hook. Allow, deny or add a note without opening the terminal. Answer in the terminal instead and the first answer wins. | | Rule-based risk labels on every request. High-risk requests are never one click and never "always allowed". | A Terminal | Review tab per session: file tree, unified or split diffs with word-level marks, rendered Markdown, line comments sent back to Claude as one review. | | A worktree per session, live git facts, overlap warnings when two sessions touch the same file, /Resume that saves the work as one commit and restores it exactly. | Context per session and 5-hour / weekly usage per account, with a pace, read from Claude Code's own status line. Multiple accounts supported. |
| <sub>Decision cards. Medium risk is one click (or A).</sub> | <sub>High risk. A force push asks you to confirm, every time.</sub> |
| <sub>Review. Changed files, split diff, βViewedβ marks that reset when Claude edits again.</sub> | <sub>βK. Jump to any session or run any action.</sub> |
Everything else it does #
- Briefing per session. Goal, what got done, your instructions and files, built from its events. No AI tokens spent.
- Calm by default. Focus windows batch everything except urgent items. Quiet hours, an hourly budget, Park everything, a daily wrap-up and "since you left".
- Stuck and crashed sessions are flagged on their own.
- Elsewhere. Claude sessions you started outside tend (another terminal, background agents) are listed
read-only. Close one there with
/exitandAdopt resumes the same conversation under tend, with cards, briefing and review. tend never stops, signals or types into a session it didn't start. - Media. Paste or drop screenshots, images, PDFs and files. They become paths at the prompt.
- Polish. Smart links, one tend per machine,
tend rotate-token, and a self-test after Claude Code updates.
npx tend-cc
That's it. tend starts and opens in your browser. Click New session, pick a folder, and you're running Claude Code from one page.
To keep it around as a tend command:
npm i -g tend-cc
tend
Needs macOS (the core works on Linux, untested), Node.js 24+, tmux and Claude Code logged in. Missing something? tend doctor tells you exactly what and how to fix it:
$ npx tend-cc doctor
β Node.js 24+ v24.4.0
β tmux β brew install tmux
β Claude Code 2.1.330 (Claude Code)
β macOS yes
Sessions live in tmux, so they keep running if tend stops. Start it again and everything comes back.
Run from source #
git clone https://github.com/Geekgineer/tend.git && cd tend
npm install
node bin/tend.mjs
| Command | What it does |
|---|---|
tend |
Start and open the page. One per machine; a second one tells you where the first is running. |
tend --no-open |
Start without opening the browser (also TEND_NO_OPEN=1 ) |
tend doctor |
Check Node, tmux and Claude Code, with the fix for anything missing |
tend rotate-token |
A new link. Old links and open tabs stop working within a second. |
TEND_PORT ,TEND_HOME |
Port (default 4977) and data folder (default ~/.tend ) |
flowchart LR
B["Your browser<br/>one page"] <-- "SSE + token" --> T["tend<br/>127.0.0.1:4977"]
T <-- "tmux control mode" --> S1["claude Β· session 1"]
T <-- "tmux control mode" --> S2["claude Β· session 2"]
T <-- "tmux control mode" --> S3["claude Β· session N"]
S1 & S2 & S3 -- "hooks + status line" --> T
T --- D[("~/.tend<br/>SQLite")]
- Every session is the real Claude Code TUI in tmux, streamed to xterm.js in the browser.
- tend writes a per-session settings file, so Claude Code reports to it through its own hooks (permissions, questions, plans, tool use) andits own status line (context, usage).
- State lives in
~/.tend(SQLite). Node.js only: no native modules and five runtime dependencies.
| π Local only | Binds to 127.0.0.1 . Every request needs the token (cookie or header), with Origin and Host checks. Only/health is open, and it carries no data. |
| β You approve, always | tend never approves anything you didn't, never passes --dangerously-skip-permissions , and never accepts Claude's trust dialog for you. |
| π Hands off your secrets | tend never reads your Claude credentials and makes no network calls of its own. |
| π No telemetry | None. Everything lives in ~/.tend . |
| Keys | Action |
|---|---|
β`` K |
Palette: jump to a session or run an action |
A /D |
Allow / deny the focused card |
Ctrl``` `` |
Switch Terminal β Review |
j /k |
Next / previous file in Review |
n /p |
Next / previous change |
v |
Mark file viewed |
β`` P |
Open any file |
Does it replace Claude Code? #
No. Every session is Claude Code itself, unmodified, in tmux. You can attach to any session from a terminal and keep working there.
Does it cost tokens? #
No. Briefings, risk labels and summaries are built from Claude Code's events by plain rules. tend spends no AI tokens of its own.
Linux? Windows? #
The core works on Linux, but the Finder folder window, notifications and image conversion don't, and Linux isn't tested. Windows isn't supported.
Multiple Claude accounts? #
Yes. tend finds your Claude config folders and shows usage per account. Pick the account when you start a session.
bin/ the tend command
src/ server: sessions, tmux, hooks, git, usage, risk, store
ui/ the page: plain ES modules, no build step
test/ unit, integration and e2e (a fake Claude, no tokens spent)
scripts/ build and lint Β· smoke/ real-Claude runs Β· probe/ browser checks Β· perf/ load and soak
docs/ design spec, per-version results, research, README assets
npm test # the full suite: fake Claude, no tokens spent
npm run typecheck
npm run check # file size, dependency allowlist, syntax
| Script | What it checks |
|---|---|
scripts/probe/e2e-browser.mjs |
The full journey in headless Chrome (19 checks, fake Claude) |
scripts/smoke/real.mjs ,decisions.mjs ,usage.mjs ,media.mjs |
Real Claude (haiku), throwaway temp repos |
scripts/smoke/package.mjs |
Pack, install into an empty folder, run |
scripts/smoke/selftest.mjs |
The Claude Code compatibility self-test (one tiny prompt) |
scripts/perf/load.mjs ,soak.mjs |
20 sessions under load; long soak runs |
Design: docs/design/2026-10-27-tend-design.md. Per-version criteria and
results: docs/versions/. More in docs/.
<sub>Built for people who run more Claude Code sessions than they have monitors.</sub>