Skip to content
ELU-LabsPublic

About

A full-screen terminal app that runs coding sessions on the AI command lines you already use, with bots that review and do work, and teammates you can ask.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

77 stars

Watchers

0 watching

Forks

Repository files navigation

ELU

A full-screen terminal app that runs coding sessions on the AI command lines you already use, with bots that review and do work, and teammates you can ask.

One message starts a session: ELU plans the change as a Task, you approve it, the agent works in its own Git worktree, ELU runs your project's checks, and the reply shows.

In the ELU terminal app, one message starts a session: ELU plans it as a Task, the agent works in its own Git worktree, the project's checks run, and the reply shows.

Install

From your home folder, not inside a project you do not trust:

npx -y @elu-labs/elu

This opens ELU and installs the elu command; next time you type elu.

You need macOS or Linux, Node 22 or newer, Git, Claude Code signed in, and Codex installed (no sign-in needed; its sandbox runs your project's checks): npm install --global @openai/codex. Using Antigravity? Read its page first. More in Install.

Start

elu

The first start asks for a project name; later starts open your last project. Type a message and press Enter. That starts a session: the model answers, or edits in a new Git worktree on its own branch.

Features

Tasks

Ask for a change and the first reply is a Task: its goal, its objectives and its success criteria, with a checklist of subtasks once the work starts. Approve it, and where keep going is on (/keepgoing on turns it on) ELU starts the session again by itself until every objective is met.

A Task in the ELU terminal app: a card with a goal and two objectives is approved with y, the session works, and ELU starts it again until both objectives are met.

Replies while it works

Write to a session while it works: the turn reads your message at its next step, and its answer shows at once under your message, with a line such as Read 12:45 PM · answered 12:46 PM. The turn keeps working.

A message written to a session of the ELU terminal app that is still working: the line under it changes to Read, then answered, and the answer shows under it while the turn keeps working.

Bots

Add a bot to a session: @reviewer reviews each change under its own name, and the agent fixes what it finds. Another bot, @lead, plans the work and hands the pieces to helpers. @orchestrator is your orchestrator: talk to it, and it runs your sessions for you, on this computer and your paired ones, with nothing to set up. /orchestrator shows how much it may do by itself.

A bot in an ELU session: @reviewer is added to a new session, reviews the agent's change under its own name, asks for one fix, and passes the change once it is fixed.

Models and agents

Change the model, the vendor or the effort in the middle of a session: /model picks another model and the next reply comes from it, in the same chat; /agent change sets a model's effort. /model auto turns on Auto model: ELU picks the model for each step, the top model to plan and the work model to edit, and each reply names the one that wrote it.

Changing the model in the middle of an ELU session: /model lists the models, the session changes from Opus to Sonnet, the next reply comes from Sonnet, and /agent change sets its effort to high.

Long jobs

Work that takes longer than a turn runs as a job. The turn ends, the job keeps running, and when it ends ELU starts the session again by itself.

A long job in ELU: the agent starts a long test run as a job and ends its turn, the job keeps running, and when it ends ELU starts the session again by itself and the agent reports the result.

Artifacts

A Task can name numbers to track. The session's Artifacts tab shows them as charts and tables, and the session fills them in while it works.

The Artifacts tab of a session in the ELU terminal app: a chart of tests passing by run starts empty and fills in while the session works, until it reaches the target.

Usage

The Usage tab lists your vendor accounts, with how much of each limit is used and how much time is left.

The ELU Usage tab lists made-up Claude and Codex accounts, each limit as a bar with the percent used and the time left, and a folded account is opened to show all its limits.

Your computers and your team

Share a project with your other computers and your teammates. Every computer on it sees the project's sessions and can write to them, and the computer that runs a session answers. They also share one Kanban board.

Two computers on one shared ELU project: a session ana starts on her laptop shows on ben's desk computer, ben writes to it from there, and the laptop's agent answers on both.

Publishing

Nothing is pushed until you type /publish, read the exact push and confirm; then ELU pushes the branch and opens a draft pull request. /permissions sets what a session may do without asking; only Full access lets it push by itself.

In the ELU terminal app, /publish shows the exact push and asks before it pushes the branch and opens a draft pull request, and /permissions sets what a session may do without asking.

More

  • Install: what you need, the vendors, the default model, the sandbox for checks.
  • Sessions: the tabs, the chat, messages while a session works, worktrees.
  • Tasks: what a Task holds, approving it, its card on the Kanban.
  • Commands and keys: the commands, schedules, elu from a shell or a session, the keys.
  • Models and Auto model: the model of a session, and letting ELU pick it for each step.
  • Long jobs: jobs that outlive a turn, and keep going.
  • Permissions: the modes, approvals, rules, MCP servers, the never list.
  • Bots: the six that come with the app, and making your own.
  • Orchestrator: the bot that runs your sessions for you, its levels, and what it learns.
  • Teammates: teams, invites, shared projects, your own computers. The wire format is in docs/peer-network.md.
  • Standup: an update from every open session.
  • Usage: accounts, limits, costs, /budget.
  • Vendors: Claude Code, Codex, Antigravity and Grok Build.
  • Security: what is trusted, and what an agent may do.

Update and remove

Quit ELU with /quit in every terminal it is open in, then run elu update. To remove ELU, follow Update and remove: it says what to stop first and what stays.

Help and contributing

Credits and license

MIT licensed. Sponsored by ELU.

@reviewer is adapted from the code-review skill in Anthropic's knowledge-work-plugins (Apache-2.0), and @qa from the qa skill in gstack by Garry Tan (MIT). The examples in test/fixtures/a2a/ come from the A2A specification (Apache-2.0). See THIRD_PARTY_NOTICES.md.

About

A full-screen terminal app that runs coding sessions on the AI command lines you already use, with bots that review and do work, and teammates you can ask.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

77 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages