Supported agents

Which coding agents Tegment integrates with, and which features each one supports.

Last updated Sep 05, 2026

Tegment doesn't ship its own agent. It runs whatever you already use. Anything that runs in a terminal works out of the box: open a terminal in a worktree, start your tool, and go. For the agents below, Tegment goes further: it can wire up the MCP server for you, launch them from a ticket with a quick action, and show their live status right in the interface.

The agents Tegment integrates with#

Tegment only offers an agent where it can find the command on your machine. An agent you have enabled but not installed stays out of the launch menus, the ticket start row and the to-do dialog, so you are never handed a launch that cannot run.

Any other terminal tool (aider, a plain shell, your own scripts) runs as an ordinary terminal program. You can point it at Tegment over MCP by adding the server by hand; live status and quick actions are reserved for the agents Tegment recognises.

Feature support#

AgentOne-click MCP setupLive status"Waiting on you" alertModel & effort per run
Antigravity CLI
Claude Code
Codex CLI
Cursor
Gemini CLI
OpenCode

What each feature means#

Model and effort per run#

Wherever you start work (an issue, a pull request, or an entry on a to-do list), the start row carries a Model & effort control beside the agent and the quick action. Closed it reads back what you chose ("Default", "Opus", "Opus and High"); open it lists the models, with the effort levels underneath.

The choice belongs to that one run and is never saved, the same way the auto-approve override works. It is baked into the terminal's restore command, so a session started on Opus resumes on Opus. Leave it on Default and Tegment passes no flag at all, which is exactly what it did before this control existed.

Claude Code is the agent that offers the picker today, listing the model families (Fable, Opus, Sonnet, Haiku) rather than pinned version ids, so the list never goes stale when a new model ships. Every other agent shows no control and launches on whatever it defaults to.

Live agent status#

For the agents that report it, every terminal shows what its agent is doing right now: thinking, running a tool, finished, or waiting on you. The state rides along on the terminal header, the tab, and the worktree row, and each project tab carries a counter of how many agents are blocked on your input, so nothing sits idle waiting for an answer you never saw.

Tegment shows what the agent is doing for every agent it recognises. The one piece that needs the agent to cooperate is the "waiting on you" signal: Tegment learns an agent has stopped to ask you something from the event it emits when it raises that question. Antigravity CLI, Claude Code, Gemini CLI, and OpenCode surface that event, so their terminals flag it automatically.