Use cases

What people use LiveShortly for, and how to set each one up. Every setup starts the same way: install the CLI, sign in, and put live in front of your agent.

Owner to confirm: the card asks for a real public session for each use case. None is linked yet; pick published sessions you're happy to feature and add their links below each section. Until then, browse the feed.

Pair with a teammate on an agent run#

Two people, one agent. One of you runs it; the other watches and steers from their own browser or terminal.

1. Start the session: live claude.
2. Invite your teammate as a commenter: live share <id> [email protected].
3. They open the link, or join from their terminal with live join <id>.
4. Their messages reach the agent at its next pause, and they can answer approvals with /allow and /deny from live join.

See Watching & commenting.

Approve risky commands from your phone#

Let an agent work through a long task while you're away from the desk, without giving it free rein.

1. Start the agent with permission control on (the default): live claude, live codex, live agy or live gemini.
2. Open the session link on your phone.
3. When the agent asks to run something, review the exact command or diff and choose Allow, Always allow for this session or Deny.

Unanswered requests are denied after 8 minutes, so a run never goes ahead unattended. To start the run from your phone as well, leave live daemon running on your machine and use New session. See Approvals and Starting sessions from the web.

Teach and learn#

Show how you actually work with an agent: the prompts, the dead ends, the fixes.

1. Make the session open (Share → Open) so students can watch without an account, or invite them by email.
2. Keep the link role at Watch for a lecture, or Watch and comment to take questions mid-run.
3. After class, the same link is the replay. Students can continue from any point with live claude --session <id>.

See Sharing & visibility and Forks & handoff.

Demos and streams#

Run a live demo where the audience follows on their own screens, and keep it as a replay afterwards.

1. Name it: live --title "friday demo" claude.
2. Publish it to the feed (Share → Publish), or use Share to 𝕏 to post the link with a preview card.
3. Read the replay before you publish: redaction hides home paths, emails and .local hostnames, not secrets.

See Publishing to the feed.

Async review of agent work#

Review what an agent did in a pull request's worth of context, not just the final diff.

1. Run the work through live and share the session with your reviewer.
2. They read the replay: each prompt, command and file change in order, with who approved what.
3. To pick the work up, they fork it: live handoff <id> prints a command that continues in the same agent from that point.

See Forks & handoff.

Keep sessions on your own infrastructure#

Teams that can't send code output to a hosted service run their own server with Docker Compose and point the CLI at it. Everything above works the same. See Self-hosting.