back to blog

The Story of Decosmic: From Cloud Agents to Local Intelligence

The Spark: A World on Fire (Nov 2022)

When OpenAI released ChatGPT on November 30, 2022, the world changed overnight. It was magical, capable of writing code, poetry, and essays in seconds. But as the initial hype settled, a critical flaw emerged: hallucinations. The model would confidently invent facts, cite non-existent legal cases, and fabricate data. It was brilliant, but untrustworthy.

The Response: Decosmic (Feb 2024)

Decosmic was born from this specific frustration. Our vision was clear: grounding AI in trusted sources.

In early 2024, we were among the first to bet on RAG (Retrieval-Augmented Generation) as the solution. The goal was to build a platform where non-technical users could easily build agents connected to their own files and integrations like Google Drive. We wanted to democratize the power of custom AI agents, making them accessible to everyone, not just engineers.

Decosmic was designed to be the "brain" that actually understood the world it operated in. Instead of relying on generic training data, it grounded agents in your real knowledge—your documents, your definitions, your processes. It promised a world without the "ML headache," where users could define what their agent should know without worrying about training pipelines or GPUs.

The Pivot: Learning from the Market

While the vision was strong, the market was moving at breakneck speed. Between late 2024 and mid-2025, we pivoted to Dense, a local-first AI meeting application.

At the time, local AI models weren't quite ready to fully realize the vision. But the philosophy of "local-first" began to take root. We realized that for AI to be truly personal and secure, it couldn't just live in the cloud—it had to live with the user. We learned that the "convenience" of the cloud often came at the cost of data sovereignty, a trade-off we were increasingly unwilling to make.

The Synthesis: Ryu

Today, we are building Ryu, the synthesis of everything we've learned.

It combines the data-grounding power of Decosmic with the local-first philosophy of Dense. Ryu is a privacy-first, on-device AI copilot that observes your context to help you work, without ever sending your sensitive data to a black box.

Philosophy: Local, Offline, Ubiquitous

Ryu is built on three core principles:

  • Local-First: Your data lives on your device. Screen observations, audio transcriptions, and conversation history never leave your machine unless you explicitly choose to share them.
  • Offline-First: You can work without internet. Your AI shouldn't stop functioning just because the network does.
  • Ubiquitous Computing: Ryu is always there, running alongside your OS. It observes your screen and listens to system audio to anticipate your needs, rather than just waiting for a command in a chat window.

    • Under the Hood

    Building on the technical foundation laid by Decosmic, Ryu is engineered for performance and privacy. It is built entirely in Rust using Tauri, rejecting the resource-heavy Electron frameworks common in other assistants. This ensures native performance, tiny binary sizes, and the security of memory safety.

    We've come a long way from that first spark in 2022, but the core mission remains the same: building meaningful tools that empower users. With Ryu, we aren't just building another chatbot; we're building an extension of your mind that respects your privacy.