remio 3.0: Our Giant Leap

Most AI tools are good at answering questions. What they struggle with is actually doing the work. You still have to find the right tool, set it up, feed it context, and supervise every step. The assistant is smart, but the burden stays with you.

That is the problem remio 3.0 is built to solve.

This week, we shipped our biggest release yet. I would go as far as saying it matters more than the original 1.0 launch itself — not because of what it adds, but because of what it changes.

For the past year and a half, we put almost everything into making remio the gold standard for frictionless digital memory. Hundreds of millions of pieces of content have been quietly preserved inside remio, and a growing number of people have let it reshape how they manage information entirely. Whenever they need to find anything, remio is the first place they trust.

Today, with 3.0, I believe remio becomes something more: the most capable and reliable co-worker in people's actual daily work. If you have ever wanted a helper that genuinely gets things done without demanding constant configuration, head to the remio website and give it a try.

3.0 is a large release, so today I want to give you an overview. I will follow up with deeper dives on specific pieces in the coming weeks.

remio: A complete Agent system

The foundation of everything in 3.0 is a fully realized Agent system. This is what takes remio from answering questions to actually doing work.

remio now has every component a modern Agent system needs: long-horizon continuous execution, a fully extensible tool suite, fine-grained context management, a multi-level memory architecture, a Skill system, MCP standard integration, on-demand software installation and operation, a coding environment, and connectivity to WeChat and Telegram. The list is genuinely comprehensive.

But an Agent is a core chip, not a finished device. What actually matters to people is a complete product they can pick up and use. Think of the original iPhone: the mobile processor was the core, but nobody fell in love with a processor. They fell in love with the phone. Everything we built on top of the Agent core is aimed at exactly that: a product that works out of the box, feels like home the moment you open it, and still delivers highly personalized service as you use it more deeply.

Defining the aApp standard

The most important thing we did in 3.0 is define what an Agentic App — an aApp — actually is, establishing a new category of AI software built natively for the agent era.

aApps are a new category of application, purpose-built for the Agent era, running on rOS, the Agent operating system embedded inside remio. rOS gives aApps a complete open framework: they can run continuously in the background, call Agent capabilities at zero marginal cost, exercise fine-grained control over Agent behavior, access a user's full data corpus, subscribe to real-time content events, build rich UI and menus using the aapp-ui standard, take over user sessions, and invoke system capabilities through a rich syscall layer.

Built on rOS and the aApp standard, we now have more than 30 apps in our aApp market. These include scheduled task runners, Smart Todo, a Q&A mode, a message gateway for WeChat and Telegram, My Headlines, AK Wiki (our implementation of Andrej Karpathy's LLM Wiki), and an academic search assistant — some from our team, some contributed by developers in the community.

These apps make the system immediately useful out of the box, while simultaneously opening up infinite expansion possibilities.

Every official app is open source. If the existing implementation does not quite fit your needs, fork it in aApp Studio and make it your own.

We will keep running developer competitions. Developers can sell their apps globally through a unified settlement system, with no platform tax. What you see today as a few dozen apps will, before long, become an ecosystem of thousands.

Skills and connectors

On the Skill side, we ship more than 20 curated official skills: PPT generation, spreadsheet analysis, Word documents, PDF workflows, deep research, Browser-Use, Scrapling, and more. We also connect to multiple global Skill Hubs, so you can search and install tens of thousands of skills on demand. Check the Skill Market aApp.

On the connector side, we have built one-click integrations for popular tools including Linear, Notion, Stripe, and Zapier. We also support the MCP standard, which opens the door to thousands of MCP-compatible services. Look for Connectors/MCP in Knowledge Base Settings, or via the + button in the chat input area.

Memory and context engineering

If Agent is the technique and aApps/Skills/MCP are the tools, then memory and context are the internal strength that makes the technique and tools actually work.

Memory has always been the weak point of Agent systems. Anyone who has used Claude Code, Cursor, or similar products has felt it: the Agent wastes significant time re-discovering basic facts and preferences, and the output frequently misses the mark. Fortunately, remio already had a strong foundation here from our digital memory work. For 3.0, we pushed it further. True to our "frictionless memory" philosophy, memory now refreshes continuously as you work. The painful experience of switching sessions and having the Agent forget what you just said is dramatically reduced.

Our five-level memory system now spans: foundational profile and general preferences, session context, daily activity logs, conversation history including chats from ChatGPT and other third-party platforms, and the full knowledge corpus. From precision to breadth, it covers everything.

Context engineering has always been a strength of remio, and we have made major improvements here too. Windows performance in particular has been completely reworked: even with tens of gigabytes of files syncing and tens of thousands of Office documents being parsed simultaneously, remio itself stays fully responsive.

Beyond performance, we added and significantly improved several context capabilities:

Recording transcription language support has been expanded with new models optimized for European and American languages. English, Spanish, German, French, and dozens more now see dramatically improved accuracy.

A system-level voice input method lets you hold Fn and speak freely inside any application. It supports LLM-based correction and polishing, and you can define a custom vocabulary dictionary.

Podcast+ now supports fetching, transcribing, and summarizing audio and video content from over a thousand platforms, including YouTube, Apple Podcasts, and Bilibili.

AI conversation sync captures your chat history from ChatGPT, Gemini, Grok, DeepSeek, and other platforms in real time, and lets you continue any of those conversations directly inside remio.

Safety: rVault

Every capability in 3.0 has a shadow concern: what if the Agent accidentally deletes or corrupts something important?

We built rVault specifically to make that fear obsolete. rVault is remio's safety layer for the Agent era. Before the Agent writes to any file or note, the system automatically creates a backup. If anything goes wrong, everything is auditable, reversible, and recoverable. For the first time, an Agent system has the safety guarantees that real-world workflows actually require.

CLI for other Agent products

Finally, we ship a CLI tool. Claude Code, Cursor, OpenClaw, Hermes Agent, and other Agent products can all call remio's CLI to perform knowledge base Q&A, search, and editing operations.

That is remio 3.0.

remio is no longer just a chat box that responds to one-off questions. It is a system with full context, callable tools, continuous execution, and safe delegation of real work. It understands your knowledge and materials. It participates in actual tasks: searching, organizing, generating, and following up. None of this requires complex configuration or manual assembly. Everything is ready the moment you install it.

We have been building toward this for a long time. We are genuinely excited it is finally here — and even more curious to see what you do with it. Come find us and let us know.