THE ASTER VISION
Give the AI you choose somewhere to work, a way to start when it is needed, and the ability to coordinate with other agents. Aster concentrates on levels 7, 8 and 9: automation, computer use and agent coordination.
This page explains our product direction. Current availability is described separately below.
A PRACTICAL FRAMEWORK
A conversation can help you think through a problem. Add context, repeatable methods and useful tools, and the kinds of work you can delegate expand. The nine levels below describe that progression in everyday terms.
These capabilities overlap. This is our educational framework, not an official ranking or a compulsory sequence. Choose the capabilities that help with your task.
THE NINE LEVELS
Open a level to explore what changes, a practical example and the benefit it can bring.
You initiate a conversation and use the response to move your thinking forward. AI can help explain a subject, draft text, compare options or reason through a problem. Choosing a more capable model may improve the response; you still decide when to ask and what to do next.
Explain this proposal in plain language and help me identify the questions I should ask.
WHAT YOU GAINA thinking and writing partner available when you need one.
Project files, saved instructions and available memory give the assistant relevant context. You can supply a business brief, a style preference or an ongoing plan so the work is less generic. What persists depends on the product and its settings; important project information should remain in an accessible record.
Use this project brief, our audience description and my writing preferences when preparing the next draft.
WHAT YOU GAINMore relevant work and less time repeating the basics.
A skill packages a way of working: instructions, examples, templates and sometimes scripts. It gives the agent a repeatable process to follow when the task fits. You can capture your own method or use a skill from a trusted source. The skill guides execution; it does not permanently retrain the underlying model.
Follow our research process: define the question, examine the sources, record uncertainty and prepare a referenced brief.
WHAT YOU GAINYour standards become a reusable part of the workflow.
Connectors, plugins and APIs provide supported access to external tools. Depending on the integration and permissions, an agent may retrieve documents, read a record or update information in a business application. A connection exposes specific capabilities; it does not automatically provide every feature of that service.
Find the latest project document in the connected workspace and update the task record with the agreed next step.
WHAT YOU GAINLess copying between a conversation and the tools where your information lives.
With suitable creation tools, AI can produce a deliverable you can inspect and refine: a slide deck, a report, an image, a chart or an animation. The useful step is turning your intent into an editable or shareable result. You provide the audience, purpose and feedback that shape the finished work.
Turn this research into a ten-slide presentation for a general audience, with clear charts and speaker notes.
WHAT YOU GAINAn idea becomes a tangible piece of work you can review and use.
Describe the behavior you want in ordinary language. A coding agent can help implement it, run checks and revise the result. You can begin without programming experience, while clear requirements, realistic scope and testing remain essential. A convincing preview is a starting point for checking that the actual behavior works.
Build a dashboard for these project milestones, then check the filtering, mobile layout and saved data.
WHAT YOU GAINA shorter path from a software idea to something you can try and improve.
You agree on a task and the condition that should start it. A schedule might prepare a weekly briefing; an event might launch a workflow when a file arrives. The system needs an available runtime, the required access and a place to deliver the result. Aster's goal is to carry that assignment through its lifecycle and make failures visible.
Every Monday morning, prepare my competitor update from current sources and save the report in my project folder.
WHAT YOU GAINRecurring work can begin without another prompt from you.
An agent needs tools that can act in the working environment. Those may include commands, file operations, browser automation or desktop controls with screen feedback. Aster Universal MCP provides a foundation for authorized computer access; broader application workflows require suitable tools and individual checks. The intended loop is to observe the state, take an action and check its effect.
Use the approved files and applications to update my spreadsheet, prepare the presentation and export the finished files.
WHAT YOU GAINUseful work happens in the environment where you already keep your files and tools.
A coordinating agent can divide a larger job, assign responsibilities and combine results. Independent tasks may run in parallel; dependent tasks need orderly handoffs. Our vision includes cooperation between a ChatGPT agent, a Claude agent and workers on connected computers, where the respective integrations support it. Aster should preserve ownership, progress and evidence throughout the assignment.
Have one agent research the topic and another prepare the deliverable, then review their work against the original requirements.
WHAT YOU GAINSpecialist help contributes to one tracked assignment and a checked final result.
WHERE ASTER CONCENTRATES
Automation determines when work starts. Computer access gives it somewhere to happen. Coordination keeps responsibilities, handoffs and results connected. Our aim is a dependable path from your instruction to useful completed work.
Automation
A supported event or schedule becomes a durable assignment with a clear goal and a destination for its result.
Computer use
Use the connected machine and software appropriate to the job, within the access you grant.
Agent coordination
Bring in additional agents when their capabilities help, with explicit responsibilities and useful handoffs.
PRODUCT VISION · CAPABILITIES DEPEND ON THE RELEASE AND INTEGRATION
SEE THE VISION IN PRACTICE
Choose a scenario to see how levels 7, 8 and 9 can fit together.
YOUR INSTRUCTION
Every Monday morning, prepare my competitor update and refresh the presentation for our team meeting.
Create the assignment with its scope, deadline and output folder.
A research agent gathers relevant changes and records where each claim came from.
A worker updates the spreadsheet and presentation using connected tools.
Check the figures, sources and requested sections, then return the files and unresolved questions.
A dated report, an updated presentation and supporting sources, with any remaining uncertainty identified.
Illustrative target workflows. Each requires compatible integrations and a system that can run the agent; these examples are not promises of current beta functionality.
WHAT GOOD COMPLETION LOOKS LIKE
Our design standard is that every delegated job should be able to answer these questions.
An identifiable agent or worker is responsible for each assignment, with clear handoffs.
Recorded actions and outputs show progress, with blockers reported explicitly.
Saved context and checkpoints support recovery after an interruption.
The requested result exists and has been checked against the agreed requirements.
THE ASTER FAMILY
Aster PA and Aster Universal MCP serve different parts of the broader direction. Their current offerings and access requirements are described on their respective pages.
YOUR ASSISTANT, BY EMAIL
A familiar way to request help through the email app you already use. The PA service has its own early-access onboarding, availability and operating boundaries.
Explore Aster PAYOUR AI, YOUR COMPUTER
Connect compatible AI clients to a computing environment you control. The published Windows and Linux technical beta supports permission-controlled commands, file transfers and host background jobs.
Downloads and setupModel Context Protocol is an open protocol for connecting AI models to external tools, data sources and systems. Aster Universal MCP uses that connection to expose supported computer tools. Schedules, desktop control and agent coordination each need their own supporting components.
CURRENT AVAILABILITY
The Windows and Linux releases provide OAuth pairing, permission-controlled commands, file transfer and background host jobs. Installation and startup are technical and manual. Follow the current release guide, including local administrator approval and the permission information reported by the client.
Read the current release and installation guideReliable event-driven assignments, application workflows, recovery and coordination across AI providers are areas of development and integration. Their availability must be verified for the specific release, client and environment. A running computer job alone does not keep an AI conversation reasoning or wake an idle agent.
Aster Universal MCP is open-source donationware under Apache License 2.0. Voluntary donations support development. AI-provider, hardware and hosting costs may be separate.
KEEP THE BIG PICTURE
A visual guide to the framework and the part of it Aster is concentrating on. Open the full-size version to read, save or share it.
No. They are overlapping capabilities, not a score or a required ladder. A well-designed recurring workflow may be all you need. Additional agents should help the task rather than add unnecessary coordination.
Compatibility depends on the client's tools, authentication, runtime and permissions. Cross-provider cooperation is part of the vision. Each integration and application workflow needs verification; the current beta does not promise universal support.
A host job can run while its computer remains available. A full AI workflow also needs a supported agent runtime, connectivity where required and any necessary approvals. Closing a chat does not by itself establish that an idle AI session can resume automatically.
The goal is to let people express useful work in ordinary language. The current Universal MCP beta still requires technical installation and host administration. The setup page explains the actual prerequisites for Windows and Linux.
Choose the machines, clients and access you authorize. The current technical beta can run powerful commands, so its permissions matter. Our coordination direction adds clear ownership, checkpoints and inspectable results; consequential actions need the appropriate authorization.
Explore the PA service if you want help through email. If you are ready to administer a host, read the Universal MCP beta's installation and release guide. Start with a bounded task whose output you can check, then expand as the workflow proves useful.
PUT A USEFUL TASK FIRST
Start with a clear goal and an outcome you can inspect. Explore the current services, or help shape the next stage of Aster.
Support the project