SuperARIS CLI
SuperARIS builds a persistent research team from Codex, Claude Code and Antigravity: a main agent, an engineer and a writer.
- 92 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Research agents run experiments and commands in the workspace, including on remote machines over SSH.
- Several agents work at once and spend limits of connected models.
Install
- Works in
Manual install
pipx install "git+https://github.com/wanshuiyin/SuperARIS.git"Then `superaris` opens the setup wizard. Python 3.10, Git and a Herdr or Orca workbench are required.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Research agents run experiments and commands in the workspace, including on remote machines over SSH.
- Several agents work at once and spend limits of connected models.
This is third-party code. Review the repository files before installing.
What it does
SuperARIS is a personal research agent built on ARIS (Auto Research In Sleep). The main agent holds the research direction and distributes work, the engineer implements experiments and checks evidence, and the writer drafts large parts of the paper. Each role can use Codex, Claude Code or Antigravity, with its own session and model. Each role gets its task, a working area and the relevant ARIS skills: research pipeline, idea discovery, experiment bridge, review loop and paper writing. The tool tracks registered tasks and result files while the main session runs, and project memory keeps goals, decisions and results in plain files and notes. It launches through the `superaris` wizard in a Herdr or Orca workbench.
Who it is for. Researchers and graduate students who carry a project from idea to paper with several agents and want to return to it the next day.
Good fit when
- A project must be carried from idea to paper: direction search, experiments, review and text.
- Experiments run on a remote machine over SSH and results must be collected and checked.
- One agent handles reasoning, another code and a third text, each needing its own session.
- The study runs for several days and goals with decisions must live in project files.
Not a fit when
- You need a short one-off task: a team of agents and a wizard are overkill.
- None of the `codex`, `claude`, `agy` agents is available with a model connected.
- The Herdr or Orca workbench is unavailable on the machine.
Example request
Run `superaris start . --host herdr`, build a team with Claude Code as main and Codex as engineer, and set the task: find weak spots in the experiment plan.Limitations
Python 3.10 or newer, Git, pipx and a Herdr workbench (default) or Orca Terminal are required. The `codex`, `claude` and `agy` agent CLIs must be set up with model access, and costs follow their plans. For human-controlled sessions follow-up works through notices, and automatic follow-up is available only to dedicated Codex sessions. SuperARIS obtains pinned ARIS resources when needed, and an existing ARIS installation can be linked. The project is early, package version 0.1.0.
How to disable. Close the workbench and uninstall the package with `pipx uninstall superaris`. Project files, notes and memory stay in your folder, so delete them by hand.
FAQ
How do I install SuperARIS?
Run `pipx install "git+https://github.com/wanshuiyin/SuperARIS.git"`, then `superaris`. Python 3.10, Git and pipx are required.
Which roles are in the team?
Three: main, engineer and writer. Each can be assigned Codex, Claude Code or Antigravity independently.
How do I start with explicit roles?
For example: `superaris start /path/to/project --role main=agy --role engineer=claude --role writer=agy`. For Orca add `--host orca`.
Which workbenches are supported?
Herdr by default and Orca optionally. Start with `superaris start . --host herdr` or `--host orca`.
Where is project memory stored?
In the existing notes and research wiki of the project. Agents update them as work goes and read them back on resume.
Discussion
No questions or reviews yet.
Related
CLI toolsARIS
ARIS (Auto-Research-In-Sleep)
A skill set for autonomous machine learning research: literature review, idea discovery, experiments and paper writing with review by a separate model
Argus runs long engineering and research tasks with four roles, where a reviewer reads results but edits nothing.
A research system on top of Codex CLI or Claude Code: runs of experiments, discussion of progress and a paper draft.
A team of 14 roles, 92 specialists and a skill library that installs into Claude Code, Codex and other agents via npx.
A skill that splits a task across twelve agents such as Codex, Claude Code and Cursor, sharing context and merging results.
ARC, a plugin for Codex and Claude Code: theoretical physics papers, citations and calculation checks.
README in short
Repository news is updated almost daily, with the latest entry on 11 November 2026. Guides live in `docs`: workbench, Orca, Windows and dependencies.