Beside your agents
Review what each agent changed, worktree by worktree.
Coding agents such as Claude Code, Codex and T3 Code work in Git worktrees: one folder and one branch per task, so several can run at once without touching your checkout. Nori lists those worktrees beside your repository, names the agent and the task behind each one, and shows what it changed, so you can review it, bring it up to date and open the pull request. Nori does not run the agents; it is the Git client you keep open next to them.
Why agents work in worktrees
A Git repository can have more than one working tree. git worktree add checks a branch out into a second folder that shares the same history, objects and remotes as the first. Each worktree has its own files, its own index and its own HEAD, so an agent can edit, build and commit in one while you keep working in another.
git worktree add ../my-app-fix-login -b fix-login
git worktree list
/Users/me/my-app 4f1c2ab [main]
/Users/me/my-app-fix-login 9e03d71 [fix-login]That is why agent tools create them: Claude Code can start a session in a new worktree, the Codex app can run a thread in one, and T3 Code can give a thread its own under ~/.t3/worktrees. Three or four tasks later, the question is no longer how to start them but what each one did, and whether it is ready to merge.
Every worktree, with the agent behind it
Nori reads the worktrees from Git, as git worktree list does, and shows each linked one in the sidebar under the repository. For each, it adds what Git does not know:
- The agent: Codex, Claude Code or T3 Code. Nori recognises a worktree from the agent’s own records on your computer, such as Claude Code’s session transcripts or the Codex threads.
- The task: the thread’s title in T3 Code and Codex, or the request that opened the session in Claude Code.
- Its size: lines added and removed, and how many commits it is ahead of and behind its base, kept current while the agent works.
The repository’s own folder is never labelled as an agent’s workspace, even when you ran an agent in it: only linked worktrees are attributed. Git stays the authority for each worktree’s path, branch and state; the agent’s name only helps you recognise it. On a computer without these agents, Nori lists the worktrees all the same.
See what it changed, then push it
Click a worktree to see everything it changed since its base: the target of its pull or merge request when it has one, otherwise the default branch of origin. Uncommitted work is included, so you see what the agent is still writing, not only what it committed. The diff is the same one Nori uses everywhere: unified or split, with syntax colour, and fast on large changes.
From there, push the branch and open its pull request on GitHub or merge request on GitLab, with the description drafted from the repository’s template. With Nori Pro, the review itself happens in the app too: read the conversation, comment on lines, follow the CI checks and merge.
To change something by hand, open the worktree in its own tab, or in your editor: Xcode, VS Code, Cursor, Zed, JetBrains IDEs and others.
Finish a worktree without losing work
Removing a worktree with git worktree remove refuses when it holds changes, and --force deletes them. Nori guides the end of a worktree instead: it can keep a detached checkout on a branch, tells you when the work is already integrated, and removes the folder while keeping the branch. Discarding a worktree that holds changes, a detached HEAD or submodules asks for an explicit confirmation, checked again at the last moment. Nori removes only what Git removes, and it can prune the registrations Git keeps for worktree folders that no longer exist.
The same agents, for small jobs in Nori
Nori can also ask an agent CLI you already have, signed in with your own account, for text: a commit message from the staged diff, an explanation of a commit or a hunk, a proposal for one conflict, a summary of a pull request for your review, or a pull request description. It works with Claude Code, Codex, Gemini CLI, Cursor or a command of your own. There is no API key to paste, and nothing passes through Nori.
For these, the agent runs in an empty private folder with its tools off: Codex, for instance, without its shell, plugins, MCP servers or web search. Each action can be switched off in Settings.
What Nori does not do
- It does not start, steer or stop agents. You run them as you do today; Nori shows their work.
- It does not create worktrees yet. Your agent tool, or
git worktree add, does that, and Nori picks the new one up. - Agent names come from Codex, Claude Code and T3 Code only. Worktrees from other tools are listed without an agent.