Zum Inhalt springen
vord Join waitlist macOS 15+ · Apple Silicon & Intel

features

stop checking every tab. vord tells you which agent needs you.

Most of these features exist for one reason: so the time your agents spend working is time you spend on something else, and you only get pulled back when a decision is actually waiting. Here's the whole set, in detail.

vord — cockpit
vord cockpit — every AI coding session in one grid, sorted by status

session hosting

vord owns every session as a real terminal

vord doesn't connect to your sessions from the outside. It hosts each one itself as a real PTY through SwiftTerm — which is why it can see every byte, and why its status detection works at all. In the same window you can run Claude Code sessions (with resume), Codex sessions, and plain shell or SSH sessions managed through a connection manager with an encrypted secret vault for your keys and credentials. Every session keeps running whether it's on screen or not. There's terminal search, and you can paste an image straight into a session with Cmd+V. Need a fresh start? Clone a repo into a new project without ever leaving vord.

  • Claude Code sessions, with resume.
  • Codex sessions.
  • Plain shell and SSH sessions, managed through a connection manager with an encrypted secret vault for your keys and credentials.
vord — cockpit
vord cockpit showing multiple sessions across projects

the status engine

the part that actually knows what each agent is doing

This is the core. Because vord hosts the PTY, it can tell working from waiting from idle without you clicking into anything. Detection runs in two layers. For Claude Code, the primary signal is hooks: the agent reports its state straight to a local HTTP listener, which is reliable and immediate. If hooks aren't available, an output heuristic fills in. Codex is tracked by tailing its rollout file as it writes. That typed wait reason is what makes the notifications precise and the remote buttons trustworthy: vord isn't guessing from scrolling text, it knows the agent is asking this specific thing. And the visibility goes deeper than the top-level session: when your agent spins up subagents, teams or workflows of its own, vord shows that hierarchy too — you see what the fleet is actually working on, not just that it's busy.

  • 🟢 working — the agent is busy, leave it alone.
  • 🟡 waiting on you — it's blocked, and vord tells you why: a permission request, a question, a plan to approve, or just finished.
  • ⚪️ idle — nothing happening.
  • limited — it hit its usage cap.
  • exited — the session ended.
vord — focus mode
vord focus view showing a session with status detection active

the cockpit

three densities, one fleet

The grid pulls every session from every project into one view, sorted with anything waiting on you at the top. Switch density to match what you're doing. There's a gallery view for the wider look, plus the small controls that keep a busy fleet honest: mark a card done, or park one to deal with later. vord also drops a peer hint into each agent that tells it how many others are working in the same project — so they're aware they're not alone in the repo.

  • Compact — pure status cards. The right view when you have a lot running and just need to scan.
  • Normal — the grid with a live terminal slice in each card, so you can read what's going on without opening anything.
  • Focus — one session at full size, fully typeable, for when you drop in to answer or steer.
vord — cockpit
vord cockpit grid showing compact, normal and focus density modes

notifications

one director, every channel, the right moment

Every status change runs through a single NotificationDirector. It builds the finished message, localizes it (English, German, French, Spanish), and fans it out to each channel at once: a macOS system notification — time-sensitive for permission prompts and questions, so they break through when they need to; a Dock badge with the count of agents waiting on you; an optional sound at the waiting moment (default is the ICQ alert — swap it or mute it); and on Pro, a push to your iPhone. Two controls keep it from becoming noise. The presence gate: if you're at your Mac, the phone push is suppressed — you get told where you actually are. And a privacy switch to hide session and project names in push, so a glance at your lock screen doesn't leak what you're building.

vord — notifications
vord notification peek showing a waiting agent alert with context

the git inspector

read the work before you trust it

An agent finishing is not the same as an agent being right. The inspector slides in from the right so you can check the diff without breaking your flow. You get light editing for the small fixes — bracket matching, multi-cursor, find and replace — but it stops there, on purpose. This is a review surface, not an editor. For real work, "Open in VS Code" or "Open in Finder" is one click away. On the branch side you can see, switch, and create branches; worktrees are shown read-only and vord never creates them. The honest limits: no rebase, merge, or blame UI. No automatic worktree management. If you need those, you need git proper — vord gets you to the point where you can trust the change, then hands off.

  • Diff tab — additions and deletions in green and red, tree-sitter highlighting, inline or side-by-side.
  • File tab — the complete file with search and a symbol outline.
  • Commit tab — stage by file or by individual hunk, with a commit message drafted by `claude -p` that you can edit before committing.
vord — inspector · diff
vord git inspector showing a unified diff with syntax highlighting

vord Pro · iPhone remote

the full cockpit, on your phone

vord Pro puts a real remote in your pocket. From your iPhone you can watch terminal output live, type, and answer permission, plan and question boxes — with native buttons, or with a typed answer of your own when the agent asks an open question. You can start, switch and steer sessions, flip the agent's permission mode without walking to your Mac, and send images from your clipboard or photo library as part of a prompt, several in one message if you need to. It works the other way too: when an agent wants to show you something, the show_to_user tool puts the image or page on your phone, with an archive you can scroll back through. There's an iOS cockpit tab so the phone isn't a stripped-down view; it's the same fleet, smaller. Approval boxes are transmitted as structured data, not parsed from raw output, so the buttons do exactly what they show. And only terminal streams ever cross the wire — your files stay on the Mac. The connection is end-to-end encrypted through a blind relay. Mac and phone each connect outward (works behind NAT), and the relay only routes ciphertext. Apple CryptoKit throughout: X25519 ECDH at pairing, HKDF-SHA256 for session keys, ChaCha20-Poly1305 AEAD on every frame, replay protection built in. Pairing is a QR scan or paste, confirmed by a six-digit fingerprint you compare on both devices.

  • End-to-end encrypted through a blind cloud relay
  • Native buttons — or your own typed answer — for every box
  • Send images, switch permission modes, start sessions
  • Pairing via QR, confirmed by six-digit fingerprint
vord on iPhone — drive your agents from anywhere
screenshot
vord — remote
screenshot
screenshot

vord Pro · session sharing · beta

share a running session with another person

The iPhone remote connects you to your own Mac. Session sharing connects someone else. You send a link or a QR code; the other person opens it in vord on their Mac — or on an iPhone with vord mobile — and watches the exact same terminal, live. It starts read-only. When you want them to take the prompt, you grant typing, and you can pull it back mid-stream. The share dies when you end it, or when the single-use link expires. The host side needs a Pro seat.

vendor-agnostic

two vendors, one fleet, managed MCP

Claude Code and Codex both run in vord, each with its own detector feeding the same status grid. You're not locked into one provider's tooling to get a unified view. The built-in MCP manager handles servers for both Claude and Codex, with live health checks, and it works only through the official CLIs — it doesn't reach behind the tools to edit their config. vord also ships its own MCP server, including a `show_to_user` tool: when an agent wants to put an image or a webpage in front of you, it appears wherever you're working, instead of getting buried in terminal output.

vord — cockpit · multi-vendor
vord cockpit with Claude Code and Codex sessions in the same grid

auto-pilot

automate the boring approvals, guard the dangerous ones

Auto-pilot is optional and off by default. When you turn it on, you write rules — a trigger, an action, a scope — for the approvals you're tired of granting by hand. There's a default protection rule that won't auto-approve anything containing destructive keywords: `rm`, `--force`, `push`, `sudo`, `delete`, `drop`. And an auto-continue rule for usage limits: when an agent hits its cap, vord parses the reset time, waits it out, and sends a single `continue` — so a session that stalled overnight is moving again by the time you're back, without you babysitting the clock.

vord — auto-pilot
vord focus mode showing the auto-pilot rule editor

optional integrations

plugs into the rest of your stack — if you want

vord works alone: no account, nothing phones home. Two integrations exist for people who run more than agents, and both are off by default.

// the full kit

everything else that keeps you in one window

stop checking every tab just to find the one that needs you

vord is in soft launch. Join the waitlist, and when your invite is ready you get in and try it free for 14 days.