AI that learnsto be yours.
A personal AI system in development. It’s being built to remember your context, learn from your feedback and help with everyday work, and to keep that progress when you switch AI models.
Join the tester poolPowerful AI.
Still explaining yourself.
You've explained your goals, your preferences and how you work. So why does every new chat feel like the first one?
New conversation
- Preferences
- Re-entered
- Goals
- Partial
- History
- Missing
New model
- Preferences
- Missing
- Goals
- Re-entered
- History
- Partial
Another platform
- Preferences
- Partial
- Goals
- Missing
- History
- Re-entered
A list of saved facts doesn't know where you left off.
MRVN OS should carry your goals, preferences and useful feedback from one task to the next, so you spend less time starting over.
Memory is only the first step.
Four connected stages show how remembered context can support learning, anticipation and useful action.
REMEMBER
Carry your goals, preferences and past decisions into the next task.
LEARN
Use feedback and past results to improve how it helps you.
ANTICIPATE
Notice an approaching deadline and work out what needs attention.
ACT
Prepare the next step, and use connected tools within the limits you set.
Practice should add up.
A correction should help with the next task, not only this one. MRVN OS is being built to keep score of what worked, what failed its own checks and what you corrected, then change how it handles that kind of work next time.
| Task | Your feedback | Next time |
|---|---|---|
| Research | Show me where that came from. | Include sources beside the claims they support. |
| Prioritising | The deadline matters more than the easy tasks. | Put deadline-sensitive work first and explain the tradeoff. |
| Writing in your style | Shorter, please. Start with the answer. | Lead with the answer and keep the supporting detail brief. |
| Planning | That schedule left no room for review. | Reserve time to check the work before the deadline. |
| Following up | One reminder is enough. | Respect your reminder preference and avoid repeated nudges. |
Examples of the intended behaviour, not measured results. MRVN OS applies feedback it has remembered; it doesn’t retrain the AI model.
- ResearchPractice: Clean source checks71 pointsLevel 4+4 Improving
- PrioritisingCorrection: Asks which deadline wins60 pointsLevel 3+7 Improving
- Writing in your styleNo change this week76 pointsLevel 4±0 Steady
- PlanningNo change this week70 pointsLevel 4±0 Steady
- Following upPractice: Reminders well timed73 pointsLevel 4+2 Improving
Prioritising +7: when deadlines clash it asks which matters more, and remembers your answer. Writing has held at 76 for three weeks with no new feedback to act on.
Illustrative example, not measured results. Points reflect your feedback and MRVN OS’s checks on its own work; the AI model itself isn’t retrained.
What it remembers depends on the task.
A deadline matters when planning a project. Your preferred tone matters when writing a reply. MRVN OS should bring forward the context each task needs.
- In focus
- Projects
- Connected
- GoalsPatternsDecisions
- Background
- ConversationsPeopleHistory
- Not needed right now
- PreferencesRoutines
In focus: Projects. Planning brings the project, its goals, past decisions and useful patterns into focus. Unrelated memories stay in the background.
- Relevant context
- The draft is due Friday. You chose a short report and prefer to plan work in the morning.
- How that changes the help
- A plan that starts with the outline tomorrow morning, leaves time for review and keeps the Friday deadline visible.
Illustration only. Memories are grouped by relevance to the selected task.
The best AI will change.
You shouldn’t have to start again.
MRVN OS is designed to keep your context separate from the AI model doing the work. Switching to another supported model shouldn’t mean rebuilding your personal history.
Change the model. Keep the progress.
Use newer capabilities for demanding work.
- Example role
- Compare project plans, reason through tradeoffs and prepare a detailed next step.
- How it connects
- A supported provider API, or subscription access where the provider explicitly permits it.
- What to consider
- Capabilities, usage caps and costs depend on the provider and connection. A chat subscription does not automatically include API access.
This comparison explains the intended architecture, not a live model connection. Supported providers, local runtimes and agent capabilities will be confirmed before testing.
Can I use my existing AI subscription?
ChatGPT offers plan usage in eligible participating apps. That does not confirm MRVN OS compatibility. Claude API access is separate from a Claude chat subscription. Any subscription-based connection must follow the provider’s supported integration and usage rules.
ChatGPT plan connections · Claude subscription and API access
A helper shouldn’t need to be asked every time.
When a deadline gets close, you shouldn’t have to remember to ask. MRVN OS is being built to notice it, prepare what it can and act only within the limits you set.
One goal.
Several steps.
Your limits.
Notice the deadline
Your project is due Friday, and the draft is unfinished.
Find the next useful step
Review the remaining work and propose time to finish the draft.
Prepare the work
Draft a checklist. Ask for approval before any action outside the limits you set.
Check the result
Record what was completed and what still needs your attention.
Example only. No real tasks are running.
Built around
one person.You.
Tell it once.
It should remember.
The aim is to carry forward the preferences you share and the feedback you give, so you spend less time repeating how you want things done.
For example, concise replies with a clear next step.
For example, planning the week before starting new work.
Past decisions, useful results and corrections you made.
Help shape
what comes next.
MRVN OS is in development. Limited alpha testing is targeted for 2027 / 2028.
ALPHA
Early testing
BETA
Wider testing
RELEASE
Planned release
Selected testers will try the system and share feedback. Earlier applications may be considered first, but selection depends on what the programme needs. A place isn’t guaranteed.
If you aren’t selected for alpha, you stay eligible for later testing.
Any tester benefits will be confirmed nearer release.
Join the tester pool.
Leave a few details to be considered for testing. The tester pool is open to people aged 18 and over.