Coding sessions

A coding session is Claude Code working on one issue. It runs on your own computer, through a small program called the runner, with your own Claude plan and your own access to the repository.

Before you start

• Claude Code is installed and signed in on your computer.
• Git on that computer can clone and push the repository.
• The repository is connected to the project in Volt: see GitHub and GitLab.

The runner works on Windows, macOS and Linux.

Connect your computer

Open Settings › Runners and press Connect runner. Volt shows one command with a code in it, for macOS and Linux or for Windows. Paste it in a terminal on the computer with your Claude Code.

The command installs the runner, connects it to your account and leaves it running in the background: from then on it starts when you sign in to the computer, and updates itself. The code works once and for a few minutes.

The page then lists the computer with the agents and repositories found on it, whether it is online, and how much of your Claude plan’s limits is used.

Only your own sessions run on your runner. Each person who delegates issues connects their own computer.

Delegate an issue

Open the issue, click Agent among its properties and choose Delegate to Claude Code. Before it starts you can set:

Setting	What it is
Runs on	Which of your computers, and whether it uses your Claude plan or an API key.
Repository	The project’s default, or another of its repositories. If the computer has no clone yet, the runner clones it with your git access first.
Model and Effort	Which Claude model works on it, and how hard it thinks: low to max.
Mode	Auto: Claude Code does what’s safe and asks you about the rest. Plan first: it shows you a plan to approve before it changes anything. Ask: it asks before every edit and command.
Note	Anything to add to the issue, like “Start with the backend”.

Your choices are remembered for the next issue.

You can also mention the agent in a comment, or let a loop hand it issues: Small bugs to Claude Code, Label “agent” to Claude Code and Weekly dependency updates are ready-made.

What happens next

1. The runner starts Claude Code in a separate git worktree of the repository, so your own checkout is untouched.
2. The session shows in Volt as it goes: what the agent reads, runs and changes.
3. When Claude Code needs a decision or wants to use a tool it may not use by itself, it asks you in Volt. Answer with Allow once, Always allow or Deny, and add a note if you like. In Plan first mode, Approve the plan or tell it what to change.
4. At the end of a turn the runner commits what was done and pushes the branch.

Volt then opens a draft pull request for the branch, linked to the issue.

Write to the session at any time to tell the agent what to do next; it reads the message at its next step. Stop ends the turn.

Follow your sessions

Sessions lists yours in groups: Needs you, Working, Ready, Waiting, Failed and Closed. A session under Needs you has a question waiting for your answer, and you get a notification when one does.

• If the computer goes offline, the session waits and goes on when the runner is back.
• If your Claude plan’s limit is reached, the session stops and goes on by itself when the limit resets.
• A teammate can open your session and read it, but only you can write to it.

When the pull request changes

Volt tells the agent what happens to its pull request: a reviewer asked for changes, a check failed, the branch has conflicts. To ask for a change yourself, comment on the diff and send the comment to the agent: see Code review.

The runner’s commands

volt-runner service status    # installed? running?
volt-runner logs -f           # what it is doing
volt-runner update            # install the latest release now
volt-runner service stop      # until the next sign-in
volt-runner repo add <path>   # work in a clone you already have

What it costs

Coding sessions spend none of the workspace’s AI credits: Claude Code runs on your own Claude plan or API key.