One agent. Here is what it does for you.
Möbius is the harness and the interface for working with AI agents: a self-hosted workspace where apps, agents, memory, skills, files, and source live together, on your computer and on your phone. Everything here is shown live at mobius.you, with the real shell and real apps.
| |
|---|
| More done, less tending | Give it a goal, schedule a task, or let it run a team of helpers while you are away. Only the decisions that need you reach you. |
| It learns how you work | It remembers what matters, reviews each day overnight, and picks up new skills. A month in, it works the way you do. |
| Apps on the fly | Describe an app. It is built beside the chat, lands on your phone, and keeps improving from your feedback. |
| Build and share, together | Share a pool of agents with the people you work with, publish apps to the App Store, and send what generalizes back to the community. |
| Connected to your tools | Notion, Linear, Google Cloud, GitLab, Stripe, Supabase, and any remote MCP service. |
| Your agent, everywhere | On the web, on your phone, on a server you control. |
Your agent starts as an intern and matures through your feedback
The agent you get on day one is a generalist. The one you have a month later knows your projects, your preferences, and the way you like things done, because everything it learns stays with it. That is where the productivity comes from: not a smarter model, a more personal one.
- Memory. Every conversation leaves something behind: a decision, a preference, how a project is wired. Your agent keeps it, links it, and cites it the next time it matters.
- Feedback that sticks. Correct it once and the correction outlives the chat. Skills, memory, and the apps it built all move with what you told it.
- Reflection. Every night it goes over the day, sharpens the skills, apps, and memory your agents work from, and leaves you a morning brief. Each change is small and reversible; together they compound.
- Skills. The playbooks behind the work, browsable and installable, sharpened by Reflection and by you.
Put it to work
Hand over whole pieces of work, not single questions.
- Goals. Type
/goal and a to-do list becomes a goal the agent keeps pursuing across turns, breaking each item into steps and checking in only when it needs you.
- Tasks. Scheduled jobs: a morning brief, a daily check, a weekly report, each a run you can read afterwards.
- Workflows and Subagents. Every run on a readable timeline: which helpers your agent created, when, and what each came back with. Delegation to Claude and Codex stays visible.
- Queue. Send a follow-up while the agent works. It waits in line, ready to edit or drop, and runs the moment the turn ends.
A usage limit parks the run and picks it back up on its own. An API key goes in through a secure input the model never sees. A voice model on your device answers out loud, and nothing leaves your Möbius. It works the same with Claude Code and with ChatGPT (Codex).
Build apps on the fly, and build them together
Say what you need in your own words: “Build a News app for the topics I follow.” “Build a simple habits app.” The agent builds it beside the conversation, opens the working result in its own pane, and keeps refining it from your feedback. On your phone the same flow ends with the app on your home screen, opening as its own app. The app you asked for is yours; the apps you like from others are one install away.
Apps can be anything: a daily news brief, a drum machine, a 3D runner, a bilingual reader with its own agent inside. Install what others built from the App Store, then publish your own. Every app is a public repository under the Möbius OS organization, so sharing an app is sharing a repo.
100x your productivity, together
Share a pool of agents with the people you work with. Contribute keeps your projects, your agents, and your decisions in one place: spin up agents on a project, review what each of them did, and have only the decisions that matter brought to you. What generalizes goes back to the community as a reviewed contribution. No autonomous rewrite ships without a person in the loop.
Yours to run. Yours to change.
Möbius is open source, MIT licensed. The agent, your apps, your memory, and the platform itself run on a server you control. Bring the provider plan you already pay for; no separate API key is needed.
Hosted for you. mobius.you creates a private deployment in a Railway account you control: sign in with Google or Apple, connect Railway, open your Möbius. Start with a $5 free trial for your agent's home; no card is required. Use the ChatGPT (Codex) or Claude Code plan you already have, or try the Möbius agent, tuned for Möbius, with $2 of free credit; Möbius subscriptions are coming soon. Your chats, files, apps, credentials, and agent activity stay inside that deployment.
On your own server. You need Docker, a domain name, and a Claude Code or ChatGPT (Codex) account:
git clone https://github.com/mobius-os/mobius.git
cd mobius
cp .env.example .env
sed -i 's/^DOMAIN=.*/DOMAIN=mobius.example.com/' .env
BUILD_SHA="$(git rev-parse HEAD)" \
BUILD_DATE="$(git show -s --format=%cs HEAD)" \
docker compose up -d --build
sudo scripts/install-rebuild-helper.sh
Caddy configures HTTPS. Open https://mobius.example.com, connect your provider, and start asking. Settings → Möbius applies platform updates; docker compose exec -u 0 app bash opens a root shell in the running container. See .env.example and ARCHITECTURE.md for the trust boundaries.
The final setup step installs the narrow host controller used by Settings →
Rebuild container. It grants only that fixed operation, not general host or
Docker access. Older self-hosted installations need to run it once from their
trusted host checkout; new installations get it in the normal setup above.
Contribute to the platform
Möbius grows through apps, platform changes, testing, and discussion. A local improvement can stay private or become a reviewed contribution through the Contribute app and GitHub. To work on the platform itself, read CONTRIBUTING.md and ARCHITECTURE.md.
License
MIT
