Glossary

Agent. An AI that writes and changes code for you: it reads files, edits them and runs commands, asking first when you want it to. Kovan has its own agent and also runs Claude Code and Codex.

AI teammate. An agent with a name and a role that is a member of the workspace, such as "Bora, reviewer". You give it work in a session, or put it in a swarm. It runs on the computer of whoever gives it work.

Approval. An agent asking "may I do this?" before an edit or a command. Who may answer depends on the role and the workspace's settings.

Brain. The workspace's shared notes: decisions, gotchas and conventions. Every agent working there reads them.

Checks. The project's own tests, linters and builds. In a swarm a task counts as done only once they pass.

Lane. One person's (or AI teammate's) place in a session: their agent's conversation, which everyone in the session sees live.

Model. The AI behind an agent, such as Claude Sonnet or GPT. Kovan's own agent can use hundreds of them.

JevRouter. Kovan's model picker: it gives each step to the cheapest model likely to get it right.

MCP server. A plug-in that gives agents more tools, such as GitHub or a database. See Plugins and MCP.

Permission mode. What an agent may do without asking: Plan only (reads and plans), Ask first (asks before every edit and command), Edit freely (edits, asks before commands), Full access (asks nothing).

Role. What a person may do in a workspace: Viewer (watches), Commenter (also comments), Editor (runs agents and changes code), Owner (also manages people and settings).

Session. One task the team works on together, on its own Git branch, with a lane for each person in it.

Swarm. Several AI teammates working on one goal: a lead splits it into tasks, the others build them side by side, and each is checked before it counts.

Workspace. A project shared with a team: the repository, its people, sessions and settings. The code stays in each person's own copy.