Glossary
Every word this documentation uses, alphabetically, in one sentence each.
Every term in this documentation, in one sentence each.
Agent. One of the command-line tools beafk runs for you — Claude
Code, Codex, Grok Build, Kimi Code. See
which agents beafk runs.
Ask. A question an agent stopped to put to you. The chat waits —
twelve hours with nobody looking — and the answer reaches it whenever you get
there. See when an agent needs you.
beafk cloud. This website: the account, the workspace, the fleet and the
bill. It holds no code and no chats.
Cabinet. beafk cloud, seen from a machine — the thing a machine reports to
and takes its permission from.
Certificate. The short-lived, device-bound permission your browser presents
to a machine to open it. Signed by us, verified by the machine on its own, and
spent at once. See how your phone reaches it.
Chat. One thread, one agent you are typing to, and up to eight folders it is
pointed at — the first is where the agent runs, the rest it may work in. It runs
in a process of its own on the machine. With a crew on,
the pieces that agent hands out are chats of their own. Not a session, which
here is what a browser holds. See a chat.
Command. A line your project is run by, kept as a button in
.beafk/commands.json. See commands.
Compress. The agent's own summarise-and-continue, for a chat whose
context has filled up.
Context meter. The gauge under the message box showing how full the agent's
memory of the chat is.
Crew. Talk to one agent. It hands pieces to the right CLIs, and brings
every result back to the same thread. See
what a crew is.
Device key. A key your browser makes and cannot export. A certificate is
issued against it, which is why a copied cookie opens nothing.
Draft. A CLI lined up on a task with the brief it should open
with, waiting for one press to start.
Effort. How hard an agent thinks about one answer. More effort is more time
and more of your subscription.
Fingerprint. The long SHA256:… name a machine calculates for itself from a
key it made. It is the reason a machine belongs to exactly one workspace.
Grant. Permission for a named person to open a particular machine, at full
or read-only level. Teams only. See
who may open what.
Hold. The claim an agent takes on the files it is about to edit, so that a
second agent queues instead of writing over the first. See
two agents, one file.
Lead. The agent you are typing to when a crew is on. It hands the pieces
out, reads what comes back, and tells you what came of it.
Lease. The signed note, good for a day and refreshed every five minutes,
that tells a machine it is paid for.
Machine. A computer you own with beafk installed. Also called a node. One
machine is one line on your bill.
MCP server. See tool server.
Node. Another word for machine, used in some screens and commands.
Pane. One real terminal on the machine, running one line you gave it, with
its output still there when you come back. See
running a line yourself.
Panel. The screen you use to drive a machine, in your browser, talking to
the machine directly. See the home screen.
Permission mode. What a turn may do to the machine without asking you first.
Each agent has its own set. See
models, effort and permission.
Project. A folder on your machine, usually a git repository. The screens —
commands, variables, git — are scoped to the one in the header; a chat can
carry several. See the home screen.
Role. A job written down: what it does, when to use it, which agent runs it,
on which model and at what effort. See
writing a role.
Seat. One connected machine on your bill. Not a person.
Session. What a browser holds: the hour a machine gives it after a
certificate, and the sign-in it holds here. Never a thread with an agent — that
is a chat. See your browsers.
Skill. A page of instructions an agent reads before a certain kind of job.
See skills.
Sketch. A clickable stub of a screen, built in minutes, so you can pick the
right one before an hour is spent. See
seeing it before it is built.
Tailnet. Your own private network, made by Tailscale, that your phone and
your machines are on.
Task. A row on a project's work list, with the chats it has been
the subject of underneath. See tasks.
Tool server (MCP). An outside service an agent can use — a ticket tracker, a
browser, a database. Described once, delivered to every agent. See
tool servers.
Trace. A record of one operation and how long it took. See
logs and timings.
Trash. Where deleted things on a machine go, with a receipt, for thirty
days. See the trash.
Usage. What the agents spent in tokens, by model — on the machine, and
across the fleet in beafk cloud. Not a dollar amount. See usage.
Workspace. Who you share machines with. A machine belongs to exactly one.
See workspaces.