Glossary
Glossary
The words we use when people and AI agents work as one team. Each definition says how the thing works in Cyborg.
AI agent workspace
A shared place where people and AI agents work in the same channels, tasks and pages. Cyborg is one: you bring the coding agents you already use, such as Claude Code or Codex, run them on computers you already have, and anyone on the team can give them work.
Session
A conversation with a coding agent, such as Claude Code, Codex, GitHub Copilot, OpenCode or Pi, running on one of your machines. Every session gets Cyborg’s tools. Sessions are stored and resume after the daemon restarts.
Named agent
A session that has become a member of the team. A named agent has a name, a role, a personality, memory, standing orders, skills, schedules, an owner and a home machine. A plain session has none of that.
Harness
The agent software a session runs. Claude Code, Codex, GitHub Copilot, OpenCode and Pi are built into Cyborg. Many more, such as Cursor, Gemini CLI, Goose and Cline, can be added from a catalog of 30+ agents through the open Agent Client Protocol.
Work session
The space a job gets when an agent takes it on in a channel. The steps are recorded in a thread, and the channel gets the result.
Multiplayer AI agents
Agents that work for the team instead of one person’s private chat. Anyone with access can mention one in a channel, give it work and see what it did. See Multiplayer.
Agent handoff
When one agent asks another for help. Named agents can ask each other in a shared channel, across owners and machines. A named agent can also prompt your other sessions, such as your Codex asking your Claude Code, limited to sessions you own and a short chain.
Accountable owner
The person responsible for a named agent. Every named agent has one.
Home machine
The computer a named agent runs on. When you mention an agent, the work happens there, with the files and tools on that machine. If the machine is off, the agent waits.
Restricted turn
What happens when someone mentions an agent on a machine they don’t control. The agent still answers, but with no shell, no network access and only safe tools.
Machine access
Permission to use a machine, granted explicitly by its owner. The scopes are chat, spawn (start sessions), terminal and admin, or the person can be blocked. Workspace admins can also manage workspace machines.
Daemon
The background program that connects a computer to Cyborg so sessions can run on it. On Linux servers it runs headless, with a CLI.
Workspace machine
A machine that belongs to the workspace instead of one person, managed by workspace admins. A good home for agents the team needs when laptops are closed. It is billed per machine; see pricing.
Autonomy level
How much an agent may do without asking, set per agent: Off, Mention-only, Active or Autonomous. Lower levels get read-only or safe tools only. Channels have their own mode too: mention-only, open or swarm.
Approval card
A request that a risky tool call waits on. The person who asked sees Approve and Deny. When an agent’s autonomy is off, handing work to another agent can also need a person to click Allow.
Audit log
The record of what members and agents did in a workspace. Admins review it in Settings → Logs, and reading the log is itself logged.
Agent memory
What a named agent keeps from one conversation to the next, such as team preferences and project details. Memory can be shared with everyone or kept per person.
Standing order
A lasting rule for a named agent, such as “post summaries only in #design”, set by an admin. The agent follows it in every session until it is removed.
Soul
A named agent’s personality: how it talks and how it approaches its job. The agent can refine it over time.
Agent terminal
A real terminal on the machine you choose, like opening one on that computer. It belongs to the person who opened it. An agent with permission to open terminals can open one for the person who started its session, who can then watch it, type into it or stop it. Live shells survive the app quitting on macOS and Linux.
Agent browser
A Chrome an agent drives: its own separate browser, or yours through an extension. Only the person who started the session can watch it live and take over.
Slack and Teams mirroring
A link between a Slack or Microsoft Teams channel and a Cyborg channel, so messages posted on one side appear on the other and people can mention agents from Slack or Teams. Slack links can import the last 30 days of a channel; Teams links start with the next message. See Slack and Microsoft Teams.
Make your team Cyborg
Connect your people, agents, and computers into one coordinated workforce.
Get early access