All your agents.
One table.
Bring Claude Code, Codex and Antigravity into one hall.
Shared skills. Connected machines. You at the head of it all.
An open-source project taking shape. First testers welcome.
A little less switching.
A lot more building.
Add pagination to the projects endpoint. Follow our API conventions.
I’ve loaded the API skill from your armory. I’ll implement the change in an isolated worktree, then run the tests.
Different tools.
The same side.
Inspired by the great halls of Norse stories, Longhall is built around a simple idea: a shared table, with an equal seat for every tool.
It brings the best working patterns from Claude Code and Codex into one place. Keep the tools you like. Give them a common way to work.
Teach it once.
Bring it everywhere.
Your best skill shouldn’t belong to one model.
Keep skills, rules and MCP servers in a central Git repository. Longhall delivers them to the tools and machines you enroll.
--- name: api-conventions description: Build consistent APIs --- # Every endpoint, the same care. Use cursor-based pagination. Validate inputs at the boundary. Cover the behavior with tests.
~/.claude~/.codex~/.geminiTry the flow. This demo changes no files.
From a good idea
to work in motion.
Start a session on your Mac. Keep work running on a server. Give recurring work a schedule. Follow it all from the same panel.
The right tool.
The right place.
A clear path
from request to review.
Good work.
On repeat.
Agents propose.
You decide.
A session finds a better way to work. It proposes a change to your armory, with a reason and a diff. You accept or reject it. Deploying it to your machines is a separate decision.
Review supported permission requests in the panel. Longhall’s dispatcher cannot answer them for you.
Write actions are audited. Armory deployment previews changes and backs up the files it manages.
A better review checklist
“We caught a missing edge case. Let’s make checking empty results part of every API review.”
## Before you finish Run the endpoint tests. + Check empty result sets. + Check pagination boundaries.
Give the work
some character.
Roles bring instructions, skills and guardrails to a session. Start with a focused builder or a chief of staff who coordinates the work.
Role plugins currently run in Claude Code.Make it. Test it.
Report back.
A focused implementation role, working in its own worktree. Finishes with a report; does not push or deploy.
Keep the bigger
picture in view.
Jarl opens sessions, delegates separate tasks and follows their progress. Coordination has a place of its own.
A small hub.
A wider world.
Run a hub on an always-on machine and an agent on each Mac or Linux host. Connect over SSH or a reverse tunnel. Open the panel in your browser.
Open the field guide
Longhall hubPanel · orders · armory · auditA few things to know.
Is Longhall open source?
Longhall is licensed under Apache 2.0. The GitHub repository is initially private while the first testers come aboard. Request an invitation below; public access is planned.
Does it replace Claude Code, Codex or Antigravity?
No. Longhall brings your existing tools into a shared workspace and distributes a common armory. You keep your own tool accounts, subscriptions and machines. It does not convert a conversation from one tool into another.
Does every tool have the same capabilities?
Not yet. Longhall can manage its own sessions across the three tools, with different controls for each. Codex threads started by other clients are read-only. Antigravity permission dialogs are answered in the terminal. Role plugins currently target Claude Code. The field guide explains the boundaries.
Can I start with just one machine?
Yes. Run the hub and an agent on the same machine, then add more hosts when useful. The current requirements are macOS or Linux, Node.js 22 or newer, tmux and the tools you want to run.
How early is early access?
Longhall is pre-1.0. Interfaces and configuration can change between minor releases. We’re inviting people who want to try real workflows, report rough edges and help shape the project.
Bring your tools.
Take a seat.
Try Longhall with your own workflow. Tell us what works, what gets in the way, and what the hall should become.
Apache 2.0 · Independent project · Early testing
github.com/marcinrabiej/longhall ↗