Use Claude Code and Codex together in one local workflow

Choose a workflow first

Start with one agent at a time unless you have a reason to work in parallel:

• Sequential review: one agent edits the project; the other reads the diff and reports issues. This is the simplest way to get a second opinion.
• Parallel implementation: each agent edits its own Git worktree and branch. You still need to review and combine the changes yourself.

Before either workflow, check git status --short. Commit or otherwise protect work you want to keep so you can review and recover the agents’ edits.

Use one agent to implement and the other to review

1. Open the repository in Lomi. On macOS or Linux, choose +, then Agents to detect an installed CLI and launch it at the project directory. You can also start a CLI from a regular terminal.
2. Ask the first agent for one focused change. Name the relevant files, expected behavior and checks. Ask for a plan first if the code is unfamiliar.
3. Wait until the first agent stops editing. Inspect the result in Lomi’s Git view or run git diff --check and git diff.
4. Start the second CLI in another terminal in the same project and give it a review-only task. Do not ask both agents to edit the same files at the same time.
5. Review each finding, make needed changes, run the project checks and inspect the final diff before committing.

For example, after Claude Code has completed an implementation, ask Codex to review the diff without changing it:

Review the current Git diff for correctness, edge cases, and missing tests.
Read the relevant code. Do not edit files or run commands that change the
working tree. Report each finding with its file and line, or say there are no
actionable issues. The task is: [describe it].

The prompt asks Codex not to edit, but check the diff yourself. If it makes changes, inspect them before continuing.

Swap the roles, or ask one agent for a plan and the other to implement it. Review the plan before handing it off.

Run Claude Code and Codex in parallel with Git worktrees

If both agents need to edit at once, create a separate working directory and branch for each. A Git worktree checks out another branch in another directory. It separates files, but does not sandbox agent processes, grant permissions or prevent conflicting changes.

1. Keep your original project open in Lomi and check that its branch and working tree are ready.
2. From the repository root, create a second worktree at a new path. Use a branch name and directory that do not already exist:

git worktree add -b codex-task ../my-project-codex HEAD
git worktree list

3. Open ../my-project-codex as a separate project in Lomi and start Codex there. Leave Claude Code in the original project, or create a second worktree for it too.
4. Tell each agent which branch, task and files it owns. Avoid overlapping changes; do not ask either agent to merge the other’s work.
5. Install the project’s dependencies in each checkout when needed. If both agents run development servers, use different ports so their local browser previews do not collide.

The example creates a branch from the current commit in a sibling directory. Use a different path or branch name if either already exists. Git worktrees share repository data; they separate files, not permissions or secrets. See Git’s worktree documentation for command details.

Review and combine the work yourself

When the agents finish, inspect each worktree’s status, uncommitted diff and any commits they created. Review each commit against your starting point and run its checks. Commit intended uncommitted edits on their own branches before merging; Git does not merge uncommitted changes.

Before merging or cherry-picking, make sure the target working tree is clean. Resolve conflicts, inspect the combined diff and run the project checks again.

Worktrees do not share uncommitted files. Install dependencies and check generated files, environment settings and local services in each directory. Use different ports for simultaneous development servers, and keep credentials out of agent prompts.

What Lomi provides, and what you set up

Lomi provides local project folders, terminals, an editor, browser previews and Git review. It does not share context between Claude Code and Codex, coordinate their tasks or create worktrees automatically.

• Install and sign in to Claude Code and Codex CLI using their official setup instructions. Their accounts, eligibility and usage charges are separate from Lomi.
• Lomi’s managed Agents launcher is available on macOS and Linux. On Windows, use a regular Lomi terminal and follow the CLI vendor’s Windows or WSL setup instructions.
• Each CLI’s permission settings still apply. Review its changes, run checks and handle the final merge yourself.

See the individual Claude Code setup guide and Codex CLI setup guide for launch details. To compare Lomi’s workflow with other coding workspaces, read the coding workspace comparison or the Warp alternative guide. Explore Lomi features or download Lomi.