On 19 November, Pulsar 2.0.0 was released to unify local coding agents across multiple computers into a single controllable pool, according to the release notes. The update keeps single-machine 1.x behavior intact by default while adding an optional web interface served at each machine's Tailscale address on port 7337. Builders can inspect live agent transcripts, review dirty git worktree diffs, and answer pending approval prompts from a phone or another computer.1
How the setup workflow runs
Builders upgrade or install the release by running npx pulsar@latest setup on each machine, which updates agent configurations and prompts whether to enable the mesh, the release notes state. Turning the mesh on with pulsar mesh on requires Tailscale with MagicDNS and HTTPS certificates enabled, or a self-hosted Headscale server using custom certificate and key files. The command installs a launchd agent on macOS or a systemd user unit on Linux, registers the Claude Code permission hook, and outputs the web app URL along with a QR code.1
How approvals and stops work
In the web app, an operator proves identity once with a passkey using Face ID, Touch ID, Windows Hello, or a device PIN, according to the release notes. Verified messages reach agents tagged with the operator's Tailscale name, while unverified or cross-machine sends carry an unverified label. An operator can halt an active turn from the app or via pulsar interrupt, and Pulsar redelivers unanswered single-recipient messages up to five times or for 24 hours across session restarts.1
Limits and breaking changes
The release notes state that Codex prompts appear as read-only cards in the app and must still be answered in the terminal. Stopping a Claude Code turn from outside fails unless the session runs inside tmux, and the daemon caps total socket connections at 1024. Additionally, pulsar stop no longer halts the background service once installed, and agents can read only their own transcripts over the socket.1