hirn architecture: local model server, model router, and desktop agent

The Assistant you actually own.

Modular. Private. Open.

A composable agent system that you own and control. Speak or type to automate your daily workflows. By default every app, model execution, and learning loop runs 100% locally.

System Layout

The 6 pillars of hirn

Each component acts as an independent service in the system, communicating locally to form a highly optimized personal intelligence node.

agent
The desktop UI and CLI agent built on top of Goose.
assistant
Cross-platform mobile companion (Android/iOS) with encrypted P2P CRDT sync.
desktop
Tauri v2 & Svelte 5 multi-agent desktop GUI and tool host for Windows, macOS, and Linux.
learn
Local finetuning loop for your personal router.
router
Intent prediction model for task and model-tier selection.
server
Efficient high-scale distributed model serving across devices.

For Developers

Ship nothing to the cloud.

Develop web Zeug, plug it into hirn's ecosystem, and forget about everything else. No servers, no deployment pipelines, no cloud bills.

Zero infrastructure

No servers to run

Zeug runs on the user's device. No backend to operate, no database to secure, no rate limits to manage.

Custom interface

Your own UI

Ship a full HTML and JS interface that opens inside hirn using a bidirectional, real-time, fully offline SDK.

Ecosystem

Plug in, not build out

Focus on what your skill actually does. Deployment, security, and data infrastructure are not a worry.

Everything stays on your local hardware. No corporate telemetry, no data collection, no exceptions. Not now, not ever. Your data, your models, your decisions.

Join the journey

Be the first to run hirn.

We are building a truly open, sovereign system. Sign up to get early developer releases and shape how hirn evolves.