Dev tools

Lody shares a running coding agent session with the rest of your team

September 5, 2026 at 4:55 PM PT

The GitHub card for the LodyAI/Lody repository

Image: GitHub

Why it mattersAgent work is currently invisible to everyone except the person who started it, so review and handover happen through screenshots instead of the actual transcript.

Lody is an open-source workspace that lets more than one person watch and drive the same coding agent session. It connects machines through a daemon and speaks the Agent Client Protocol, so it works with agents that already support that standard rather than wrapping any one of them. The GitHub API showed 886 stars, 102 forks and an Apache-2.0 licence when checked on 5 September 2026, against a repository created on 7 August.

The README names Claude Code, Codex, Kimi and OpenCode as supported, along with other ACP-compatible agents. ACP is the open standard Zed Industries published for connecting coding agents to editors, using JSON-RPC over a subprocess. Building on it means Lody inherits new agents as they adopt the protocol instead of maintaining an adapter for each.

The problem it names

Right now an agent session lives in one terminal on one laptop. Everyone else finds out what it did through a screenshot, a pasted log, or the pull request that turns up afterwards. Lody's claim is that a teammate can open the same conversation transcript, the runtime status, the files and the code changes, from a different device, while it is still running.

Dispatch works in both directions. Work can be started from desktop, mobile, web or the CLI, and the session stays visible on all of them. The iOS client uses Live Activities, so a long agent run surfaces on the lock screen rather than needing the app open.

Worktrees are the interesting design choice

Sessions can each get their own Git worktree. The README's stated reason is so agents can work in parallel without mixing changes into each other.

That detail matters more than the sharing does. Running two agents against one checkout is the standard way to lose an afternoon, because neither one knows the other is editing. A worktree per session makes parallel agent work a normal thing to attempt rather than something you do carefully once and then stop doing.

Underneath, state is synchronised with CRDTs through the Loro Stack, which is what makes two people looking at one session workable without a central lock. The README says Lody is still moving toward full local-first support, and lists documents and document sandboxes as planned rather than built.

The part to check before you trust it

Machines connect by running a daemon, and the README says they stay private until the owner shares them. It does not describe the authentication or the security model in any detail, and that is the gap worth closing before pointing this at a repository that matters. A tool whose whole purpose is giving other people a live view of a process that can edit files and run terminal commands is a tool where the sharing boundary is the entire product.

The shape of the problem is real regardless of whether this specific project is the answer. As more of the work moves into agent sessions, the record of how something got built stops being the commit history and starts being a transcript that only one person ever saw. A team that cannot read each other's sessions is a team where review happens strictly after the fact, on the diff, with no view of the reasoning that produced it. Whether the fix is a shared workspace or better session logging, the gap is going to need filling.

Source

LodyAI/Lody, GitHub. Star and licence figures from the GitHub API, checked 5 September 2026.

Source: GitHub

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

More from AI News

Codenotch puts your Claude, Cursor and Codex usage limits on the edge of the screen

Codenotch is a macOS app that pins how much of each coding agent's session limit you have used to the edge of the screen, and it reached 196 stars within hours of being published.

Dev tools

Ponytail makes a coding agent check seven things before it writes any new code

Ponytail is an MIT-licensed rule set that makes a coding agent work through a seven-step ladder of cheaper options before it writes new code, and it has reached 127,785 stars on GitHub.

Dev tools

A Codex and Claude Code hook blocks AI agents from adding checksums and disclaimers nobody asked for

The lennney/stop-that-shit repository ships a set of prompt skills and PreToolUse hooks that block AI coding agents from producing SHA-256 checksums, hardening layers, and hedging sentences the user never requested, and it has picked up 1,497 stars in the 24 days since it was created.

Dev tools