OtherU builds software for running AI on your own hardware. Ember provides local DeepSeek-V4-Flash inference on AMD Strix Halo. Hermes Agent can connect that model to tools, persistent memory, and messaging channels.
Start here
| What you want to do | Read |
|---|---|
| Understand the system | Architecture Overview |
| Run Ember | Setup and Deployment |
| Connect an application | API and Integrations |
| Add an agent or messaging channel | Connect Hermes Agent |
| Understand access and data handling | Security Model |
| Diagnose a deployment | Operations Runbook |
Reference platform
The reference machine uses an AMD Ryzen AI Max+ 395 with Radeon 8060S graphics and 128 GB unified memory, running native Linux. The supported Ember deployment uses Docker Compose. A hardware KVM is optional and is only needed for controlling a separate computer.
The standard Compose endpoint is http://127.0.0.1:8080/v1. Existing installations may use a different port; OtherU's internal inference host uses 8000. Examples in these docs use the standard Compose port.
Local inference keeps model execution on the host. Data sent through messaging channels, external tools, or cloud-backed integrations follows the configuration of those services.
Project and licensing
See Project Status for the released components and current scope. Ember is MIT-licensed; agent software, models, and legacy OtherU Core releases have their own terms. License explains where to find the applicable license.
For general inquiries, contact otheru@otheru.ai. Report vulnerabilities to security@otheru.ai.