What is OtherU's inference engine?
Ember runs DeepSeek-V4-Flash on AMD Strix Halo. Hermes Agent can connect it to tools and messaging channels.
What hardware is supported?
The standard Ember deployment targets native Linux on Strix Halo (gfx1151) with approximately 128 GiB unified memory. It is not a general-purpose GPU runtime. See Hardware.
How do I install it?
Use the release's Docker Compose configuration and preflight script. Setup and Deployment covers startup and verification.
Can I call it from my own application?
Yes. Ember provides compatible chat, responses, messages, and completions interfaces. The standard local base URL is http://127.0.0.1:8080/v1; see API and Integrations.
Does all data stay on my machine?
Model inference runs locally. Data may leave the host through configured messaging channels, external tools, memory-model providers, or other integrations. Local inference alone does not create an air gap. See Security Model.
Can it control another computer?
A configured agent can use a hardware KVM to capture the display and deliver keyboard and mouse input without installing an agent in the target OS. That access carries the permissions of the target session. See Hardware.
How does memory work?
Conversation history, persistent memory, and Ember's inference cache serve different purposes. Hermes Agent and Hindsight Memory explain the agent-side stores.
Is it production-ready?
Ember is released and actively developed. Evaluate the selected version, workload, and operational requirements; internal deployment is not a guarantee of stability or availability. Project Status describes current scope.
Can I use Ember commercially?
Ember is MIT-licensed. Models and integrations have their own terms; see License.
How do I get in touch?
Email otheru@otheru.ai for general inquiries and security@otheru.ai for vulnerability reports.