The AROS Suite
AROS is in prelaunch development. Explore current software and curated package previews; commercial checkout is not open. Each component has its own implementation and release status.
AROS Core
Technical previewconnect your agent with relevant knowledge
The memory and retrieval foundation for preserving working context and finding relevant experience. Supported integrations and behavior depend on the installed release.
Capture, memory, retrieval and candidate-improvement features have source implementations and release records. Proposed improvements are evaluated before promotion; this does not by itself show better scientific outcomes. Download availability depends on the tested platform release and your workspace access.
These installers are not yet code-signed
Windows SmartScreen and macOS Gatekeeper will warn on first launch. Verify the SHA-256 shown after download against what you expect before running it.
AROS IDE
In developmentinspect and correct
A Theia-based interface for inspecting context, task execution and outputs, and correcting research workflows connected to Core.
In active development, with no general IDE download available. The first supported workflow pilot does not require a new IDE.
No build to download yet
A tested IDE release is not available to download through this page.
AROS Commander
Early — waitlistcoordinate lab machines
A planned coordination layer for work across lab workstations and instrument-attached hosts, developed through the AROS-agent integration work. Fleet coordination remains an early development objective. Current integration code does not demonstrate a tested product for coordinating multiple machines.
Not priced before launch, and checkout is not open.
Join the waitlistPlanned Core integration
The knowledge your agent needs, when it needs it.
Core has existing memory and retrieval implementations. The complete discovery workflow described here, including local model selection and coverage across tools, remains in development. Each release’s records describe its tested capabilities and limitations.
The planned Core integration is designed to help your AI agent discover relevant tools and receive the skills and scientific context needed for the task. It aims to reduce manual context management by revealing useful detail as the work progresses.
In this design, each tool would provide a versioned set of skills and scientific context. Core would organize these resources by field and workflow similarity, with cross-links for shared methods. A small local language model would help interpret requests and select relevant material for the user’s agent.
01
Discover
Start with compact descriptions of candidate tools and workflows.
02
Select
Review inputs, scientific assumptions, requirements and what has been verified.
03
Execute
Load the selected operation’s instructions, supported parameters and checks.
04
Investigate
Open detailed methods, code, references and troubleshooting when needed.
This planned flow follows progressive disclosure: supply useful detail as the task develops. Essential constraints must be available before action. The integration must let researchers inspect the selected context and its sources, and must check permissions, versions and input compatibility in software.
Read the design, tested capabilities and limitationsDated reference-instance records
These records describe activity and stored artifacts, not customer benefit or learning efficacy. Measured on the reference instance as of 7 September 2026 — one lab's brain, not a federation-wide total.
- World facts
- 22,714
- Mental models retained
- 64,964
- Skill-hierarchy rows
- 2,197
- Lifecycle events
- 1,037