Helmsman uses a small, consistent vocabulary. This page defines the words that recur across these guides, the changelog, and the app itself, so a term means the same thing everywhere you meet it. It is for anyone who wants the plain meaning of a word before reading the guide that uses it.
The session grid
The session grid is Helmsman’s main screen. It lays out your running Claude Code sessions four across and three down, so up to twelve of them stay on screen at once. That screen is a deck. Every slot in it is a cell, and every cell holds one session.
You work from the grid. You scan it to see which sessions are busy and which are waiting, then move into the one that needs you. Because every cell is visible, there is no tab bar to click through and no window hidden behind another; the whole set is in view.
Twelve is the number that fits on one deck without the grid turning into noise, not the number of sessions you can run. A thirteenth session opens a second deck and keeps running there, so you grow past a dozen and move between decks. Fifty sessions in all is where Helmsman stops.
The grid is the primary surface in Helmsman, and the rest of the app is built around it: you spawn into it, focus a cell inside it, steer, then step back to it. See The session grid for the full picture.
Cell
A cell is one slot in the session grid. Each cell holds a single Claude Code session and shows that session’s live terminal, so a cell is where you read what an agent is doing and type back to it.
A cell is in one of three states. An empty cell is a free slot, ready to receive a new session. A live cell holds a running session you can read and type into. The focused cell is the live cell you are steering right now, the one your keystrokes go to.
A cell is a real, interactive terminal, not a read-only log. Anything you would do in your own
terminal, you can do in a cell: answer a prompt, paste a path, refine an instruction. It behaves
exactly like the Claude Code you run yourself, because it is the same claude binary, embedded
in the grid.
You open a cell by spawning a session into it, and you free it again by closing the session when you are done. There are twelve cells to a deck. When they are all full, close one to free a slot or move to the next deck.
Session
A session is one running Claude Code process, living inside a cell. When you spawn a session,
Helmsman launches a fresh claude in a real terminal, with the working directory set to the
project you picked. From there it behaves like the Claude Code you already run: you type prompts,
it works, you answer its questions.
One session fills one cell, so a deck of the grid shows up to twelve sessions at once.
Each session is independent. It has its own terminal, its own project, and its own place in the grid, and it does not share input with its neighbors. Whatever you type goes only to the focused session.
A session runs until you close it. Closing a session ends that claude process and frees its
cell for the next one. A session that looks stuck is often just waiting for an answer: focus its
cell and reply the way you would in your terminal. If it is genuinely hung, close it and spawn a
fresh one.
Spawn
To spawn is to start a new Claude Code session in a cell. You pick an empty cell and spawn into
it; Helmsman launches a fresh claude in a real terminal, with the working directory set to the
project you chose. A cell can only spawn a session if claude is installed and on your path,
because Helmsman drives your own CLI rather than bundling one.
Project folder
Your projects folder is the folder on your Mac that holds your repositories. Helmsman asks for it on first run and remembers it, so you choose it only once. When you spawn a session it starts in the project you picked, so each cell works inside a real repo. If an agent is working in the wrong place, close the cell and spawn a new session pointed at the right project.
Focus
To focus a cell is to bring it to your attention so your keystrokes go to that session. Only the focused cell receives input, so the other sessions never steal what you type. You focus a cell to do something in it: type a prompt, approve an action, paste a path. When the session is moving again, you step back to the grid and let it run.
Steering
Steering is directing a session: answering its question, refining the prompt, or pointing the agent at the next file. Because a cell is a real Claude Code session, steering is just using the terminal. Anything you would type in your own terminal works the same way here.
Steering is one beat in the working rhythm the grid is built around: scan, focus, steer, step back. You scan the grid to find the session that needs you. You focus its cell so your keystrokes land there. You steer, giving the answer or the next instruction. Then you step back to the grid and let the session run while your attention returns to the whole fleet.
Steering is not babysitting one agent. The point of the grid is that you steer whichever session needs a hand at that moment, then move on, so a dozen agents stay productive from one window instead of fifteen scattered tabs. The word covers small corrections as much as large ones: a yes or no at a prompt is steering, and so is redirecting an agent onto a different task.
See Focus and steer for the loop in full.
Deck
A deck is one screen of the grid, holding up to twelve cells; it is sometimes called a page. Twelve is what stays glanceable at once, so it is the deck size by design, not a limit on how many sessions you can run. Spin up more than twelve and the extra sessions keep running on the next deck, which the rail along the edge names and switches between. If every cell on a deck is full, close one to free a slot or move to the next deck. Across every deck together, Helmsman runs up to fifty sessions; the fifty-first spawn is refused until you close something.
Agent fleet
An agent fleet, or just the fleet, is the whole set of sessions you have running at once. Where a single session is one agent in one cell, the fleet is all of them together, spread across your projects and working in parallel while you supervise from the grid.
The fleet is why the grid exists. You spin up sessions across different repositories, let them run, and keep the whole group in view. A quick scan tells you which agents are producing output and which are waiting at a prompt, without reading any of them line by line. Your attention moves to whichever one needs it, then returns to the fleet as a whole.
Running a fleet this way replaces the old pattern of many terminal tabs and windows, where each agent hides behind the last and there is no overview of what any of them is doing. In Helmsman the fleet is one calm grid you steer from a single window.
A fleet is up to fifty sessions on one Mac, on every tier: the session grid is free and ungated. Running a fleet across more than one machine belongs to the paid Pro tier. See Focus and steer for how you keep a fleet productive.
Fleet MCP
Fleet MCP is a Pro feature that lets your agents steer the grid themselves. MCP, the Model Context Protocol, is an open standard for giving an agent tools it can call; Fleet MCP gives an agent tools for the fleet, so a running session can drive the grid much as you do from the keyboard. On the pricing page it is listed as part of Pro, alongside running across more than one machine. It is not part of the free tier or the bring-your-own-key lane.
The prompt enhancer
The prompt enhancer is an AI feature on its way into the app: it rewrites a rough prompt into a clearer one. You select what you typed and press ⌘E; a model rewrites it into the prompt you meant, naming the file, the symptom, and a test where it can. The plan the site describes: on a capable Mac it runs locally and your prompt never leaves the machine, and with your own Anthropic key, or on Pro, it can use a cloud model instead. The homepage shows it as coming soon.
Auto-update
Auto-update is how Helmsman keeps itself current. When a newer beta ships, the app offers it from a chip in the titlebar, downloads it in the background, and a restart lands you on the new version, so you download from the members page only once. The update feed is private and invite-gated: only installs tied to an active invite can see new builds. Update problems leave a trace in the app log rather than failing silently.
Private beta and invite
Helmsman is in a private, invite-only beta. You request access through the access flow, and when a spot opens Helmsman emails your invite. Once your invite is active, the download page holds the current build for your account, as a gated download tied to the email you signed in with. The same invite gates the update feed, so only invited installs receive new builds.