Start Your First Session

Choose a base session or worktree, pick the right backend, and send the first message.

Once a project exists, Jean gives you two main ways to begin.

Choose a starting point

Use a base session when you want to work directly in the project folder. This is a good fit for repo exploration, small changes, planning, and review work.

Use a worktree when you want a separate branch and path for the task. This is a good fit for feature work, bug fixes, PR reviews, and parallel tasks.

Create a worktree from real context

The new-session modal is more than a branch creator.

You can start from:

quick actions
GitHub issues
pull requests
security alerts and advisories
existing branches
Linear issues

For many of those entry points, Jean can create the workspace and immediately start an investigation flow with the relevant context already loaded.

Pick a backend and model

Each session can carry its own choices for:

backend
model
provider
reasoning or thinking level
execution mode

That means one worktree can hold several sessions with different personalities or roles.

Send the first message

In the chat window you can:

type normally
mention files
attach images
attach text files
include skills or extra context

Jean keeps that session tied to the current workspace so the conversation does not drift away from the branch you are working on.

File mentions and slash commands

Two popovers speed up input:

@ opens a fuzzy file picker. Type a name, press Enter, and the file is attached as a mention.
/ opens the slash menu. It surfaces backend commands and installed plugin skills (for example, Claude Code skills or Codex skills). Up to 15 results with arrow-key navigation.

Both popovers are keyboard-first so you can stay in the input while adding context.

Pasting content

Jean processes pasted images and large blocks of text:

images pasted from the clipboard are resized and compressed before being attached (max 1568px on the longest side). Animated GIFs and small images are left alone.
text over roughly 2000 characters is automatically saved as an attached text file instead of filling the input.

That keeps token usage low and the input readable.

Execution modes

Jean exposes three familiar modes for work that needs different levels of autonomy:

Plan for careful, approval-oriented work
Build for editing-heavy work with guardrails
Yolo for fast, fully trusted execution

Use the stricter mode until you trust the task, the repo, and the backend behavior.

Shift+Tab cycles execution mode inside the chat. Cursor Agent only supports Plan and Yolo.

Plan approvals

When Plan mode returns a plan, Jean shows it in a dialog before any execution. The buttons let you:

Approve and run the plan in the current session
Approve and clear context to start fresh with only the plan as input
Approve in new worktree to execute the plan in an isolated branch and path

Those two extra options are useful when a plan grew out of exploration but deserves a clean execution environment.

Session tabs

A single worktree can hold multiple session tabs.

That is useful when you want to split work such as:

implementation vs review
investigation vs fix
release prep vs active coding
one backend for planning and another for execution