SuperARIS

SuperARIS — Your personal research agent with ARIS (Auto Research In Sleep, ~20k stars), powered by a team of agents.

▶ Watch the full 1080p demo · ≈21 seconds

Animated preview · click to play the full video. Illustrative project and data.

Orca multi-model team: Interactive homepage · English video · 中文视频. The single-model team (3 Codex agents) and its original English/中文 videos remain available.

Quick Start

Have Python 3.10+, Git and pipx ready, plus your chosen agent CLIs (codex, claude, agy) with model access configured. Install once, then open the wizard:

pipx install "git+https://github.com/wanshuiyin/SuperARIS.git"
superaris

Choose your language, project, workbench, team and models. Existing Herdr or Orca installations are reused.

For an explicit start, open a new project's directory and choose one workbench:

For Orca, install and open the official app first; if its CLI is unavailable, register it in Settings → General → Orca CLI. Reopening an existing project keeps its recorded workbench and team. More installation options ↓

Discuss a research goal with your personal agent. It organizes a persistent team, delegates work, checks incoming results, and keeps track of what should happen next. Join any teammate's session to explore an idea, refine a draft, or take over.

Built on ARIS (Auto Research In Sleep, ~20k GitHub stars), SuperARIS equips each role with the research workflows and project context it needs. The team works within one project, with clear responsibilities and a record you can return to tomorrow.

One personal agent, a persistent team

Your main agent holds the research direction and coordinates the work. Codex can implement experiments and check evidence; Gemini can discuss the argument and write substantial paper sections. The main role can also execute or code directly within its authorized scope. Each collaborator has its own ongoing session, so you can work directly with the teammate you need and continue the discussion where you left off.

Every new team has three roles: main, engineer and writer. Recommended presets offer an all-Codex team, Claude Code main/writer with Codex engineering, or Claude Code / Codex / Antigravity. Choose Custom team in the wizard to assign Codex, Claude Code or Antigravity to each role independently, including an AGY main and teams without Codex. Each role keeps its own model choice and session. Temporary MCP workers and fresh reviewers remain available alongside the persistent team.

ARIS for every teammate

Each role receives its task, working area, ARIS navigation, and the relevant skills to read before acting. Ask the main agent to run a complete pipeline, delegate a connected workflow, or use one skill for a focused task.

These use your configured tools and keep ARIS's review and acceptance criteria. Skills and project context →

Follow up on the work

SuperARIS tracks registered tasks, result files, and reports while the original main session is running—even after you close the workbench window. Your main agent uses those updates to collect finished work, check the evidence, and continue the next authorized step.

Human-controlled sessions receive notices; dedicated Codex sessions can use automatic follow-up. Receiving results and continuing work →

How the team, receiver, main agent and project records connect (Chinese labels). Click to view full size.

Long-term project memory

Keep the project's goals, decisions, corrections, results, and next actions in its existing notes and research wiki. Agents update these records as the work progresses and read them back when resuming. The context stays with the project across days and sessions, in files you can inspect and edit.

Reopen a running team to continue its conversations, or give a new session the project's recovery entry. Context and recovery →

Project files carry progress into later sessions (Chinese labels). Click to view full size.

Install and open

Requires Python 3.10+ on macOS, Linux or Windows, Herdr (default) or optional Orca Terminal v1, and your chosen native CLIs (codex, claude, agy) with model access configured.

pipx install 'git+https://github.com/wanshuiyin/SuperARIS.git'
superaris

Choose the interface language, project, team, and models, then confirm the workspace. Give the main agent a task, or open the team and discuss what to do next. The interface language is independent of your paper or task language.

The presets are shortcuts, not role restrictions. For example, choose AGY main, CC engineering and AGY writing in Custom team, or use:

superaris start /path/to/project --role main=agy --role engineer=claude --role writer=agy

Add --host orca for a new Orca project, or --model ROLE=MODEL for individual model requests. The two AGY roles have separate sessions. CC/AGY follow-up remains notice-based; verified dedicated Codex automation keeps its existing input guards.

New workbenches use the same three-pane layout. Existing teams and pending startup choices stay intact; --only-role selects an explicit subset.

SuperARIS obtains its pinned ARIS resources when needed. Existing ARIS installations and custom skills can be used through the documented source selection. Dependencies and updates →

Local wheel (including offline installation):

python3 -m pip install --no-index --no-deps /path/to/superaris-0.1.0-py3-none-any.whl
python3 -m superaris

Checkout in an isolated environment:

git clone https://github.com/wanshuiyin/SuperARIS.git
cd SuperARIS
python3 -m venv .venv
.venv/bin/python -m pip install .
.venv/bin/superaris

Select an existing ARIS source explicitly:

SUPERARIS_ARIS_REPO=/path/to/aris superaris

Help, plans and wizard cancellation do not fetch ARIS. See [source selection](docs/DEPENDENCIES.md