OtherU's released inference software is Ember. Agent, memory, and hardware-control integrations are configured separately. This page describes the available components without treating internal use as a stability guarantee.
Available components
| Component | Scope |
|---|---|
| Ember | Released C/C++ inference engine for DeepSeek-V4-Flash on AMD Strix Halo; actively developed |
| Vision model support | Experimental DeepSeek-V4-Flash-Vision-Exp support in the default deployment since 2026.9.3 |
| Persistent KV cache | Reuses eligible model context; separate from conversation and long-term memory storage |
| Hermes Agent integration | Connects an operator-managed Hermes installation to Ember's compatible API |
| Hindsight integration | Optional persistent memory provider for Hermes |
| Hardware KVM integration | Optional capture and desktop input through configured agent tools |
The supported Ember platform is native Linux on Strix Halo (gfx1151) with approximately 128 GiB unified memory. See Hardware and Setup and Deployment.
Release expectations
Review Ember releases for changes and select a known version for a deployment. Model artifacts and images are versioned together by the release configuration. Interface support and experimental features should be assessed against that version.
Performance measurements depend on hardware, prompt length, output, and cache state. Use the performance dashboard for dated results and methodology.
Earlier architecture and product concepts
The earlier multi-service OtherU Core stack has been retired from the reference deployment. OtherU Link, Vault, and Eye were described as future concepts in earlier material; they are not installation requirements or capabilities of the released Ember software. This documentation does not assign delivery dates to them.