The words crewkit uses, each defined once, with where it appears on the dashboard and which page explains it. Where a thing has two names today, its entry gives both.
A prompt file for Claude Code that crewkit keeps and syncs to .claude/agents/, so every developer runs the same configuration. Agents are one kind of resource, with skills, commands and playbooks. Not the same as a Team Agent or an agent role.
On the dashboard:Agents in the sidebar, and a project's Agents tab.
More:Agents
A setting on each member that changes how agents work with them: coaching, collaborative or autonomous. It is separate from the member's permission level.
The organization setting that decides how much session content crewkit collects: minimal, structured or full. An organization admin sets it, for every member. An organization that never set it is on structured.
On the dashboard:Settings → Organization → Privacy & Telemetry.
More:Capture modes
A fixed copy, made when a person confirms an Ask, of its questions, its Decisions and the Responses that person was shown. Agents read only the latest copy. Respondents appear in it numbered, never named.
A conflict crewkit found between a new document and an earlier one in the same project, by sending both to an Anthropic model. A person reviews each one.
On the dashboard: a project's Contradictions tab.
More: no page of its own yet; see Documents for the documents it compares.
A conversation is the stored record of one piece of work with Claude Code; a session is the running Claude Code process that produces it. The dashboard lists conversations inside a project. The CLI command and the API path that handle the record are named sessions, for the process that starts it. A conversation holds what the capture mode collects, and members read it as described under Viewer.
A data plane is a crewkit service an organization runs in its own cloud to hold its session content. An organization that uses one is resident: the CLI sends transcript and hook-event content to the data plane. crewkit stores no transcript text, first prompt or summary for it; it receives reduced hook events and each session's metrics: who ran it, the project, when, tokens and cost.
On the dashboard:Settings → Organization, the Data Plane card.
More:Capture modes
The answer a person records for one question of an Ask, after reading its Responses, with the reason. A majority of Responses never makes a Decision on its own.
On the dashboard: on the Ask, in a project's Asks tab.
More:Asks
A project's memory: uploaded files, Google Drive imports, captured email and Slack messages, and past sessions, all searchable. Captured email and Slack messages can be found by search, by members and by agents, as soon as they arrive; crewkit adds them to a session's context on its own only after a person confirms them. A document's text is sent to Voyage AI to make it searchable and to Anthropic for analysis, except where your organization's data settings bar it. The dashboard says Documents; the API calls them Artifacts.
On the dashboard: a project's Documents tab.
More:Documents
Two different things. Feedback is the panel a team adds to its own app, so the people using that app can send the team a message. A report also carries the name and email they typed, the page, the browser and an optional screenshot. A report nobody confirms is deleted 30 days after it arrives; a confirmed one is kept 180 days by default. crewkit feedback "message" is a CLI command that sends your own message about crewkit to the crewkit team.
On the dashboard: a project's Feedback tab, and pending reports in the Inbox.
More:Feedback, CLI
The page where content waits for a person to confirm it: Asks with Responses not yet confirmed, inbound messages crewkit could not file to a project, code-host items to confirm, email and Slack documents held until a person confirms them for agents, and pending Feedback. Email and Slack messages are searchable while they wait.
A short rule a person on your team wrote or approved after the same evidence came up for two or more people on a project. Accepted Lessons are added to later sessions on that project. Lessons are off until crewkit has enabled them for your organization and an organization admin has switched them on.
On the dashboard: a project's Lessons tab.
More:Lessons
A playbook is a collection of conventions: the team's standards, such as which test framework to use. Your organization subscribes to playbooks and assigns them to projects.
On the dashboard:Playbooks in the sidebar, and a project's Playbooks tab. Analytics → Conventions shows when sessions overrode a convention.
More:Playbooks
A project is where crewkit keeps a team's context, conversations and settings. A workspace is a kind of project that runs work across several repository projects.
On the dashboard:Projects in the sidebar. A new project is a repository project or a Workspace.
More:Workspaces
A session started with crewkit --sensitive. In an organization crewkit hosts, its transcript and hook events are still sent and stored under the capture mode, but only the person who ran it can find or open it, not a manager or an admin. It is left out of session analysis, project search, correction detection, convention override resolution and Lessons, and only that person can share it.
On the dashboard: none; you choose it when you start the session.
More:CLI, Capture modes
Messaging between your Claude Code sessions that are already running, on any machine. Only the messages each session sends cross; no conversation, context or transcript is shared. Each message is kept in both conversations' transcripts, and is readable by whoever can read those conversations.
On the dashboard:Settings → Session links.
More:Session links
A link that shows one conversation at the level you chose: to signed-in members of your organization or, when an admin makes it Public, to anyone holding the link, with no account. Signed-in members can add annotations to it. It expires when you set, or never.
On the dashboard:Shares in the sidebar.
More:Session sharing
A project's agent teammates, each with its own email address and an optional Slack handle. The dashboard says Team Agents; the docs page calls them agent identities. Not the same as an Agent.
On the dashboard: a project's Team Agents tab.
More:Agent identities
The read-only permission level. A Viewer reads analytics, documents, Asks and other members' conversations, transcripts included. It cannot read conversations in Restricted or Private repositories, or sensitive sessions. The content of a session that took in a meeting transcript is read by managers and above. What a Viewer costs is on the pricing page.