Is OtherU production-ready?
OtherU Core is in active development on dedicated hardware. The multi-agent system — coordinator, responder, reasoner, coder, and Fara — all are functional. Serious test and evaluation needs to be completed before considered production ready. Some components (latent-mas v3, OtherU Link/Vault/Eye) are research-stage or future modules.
What hardware does OtherU run on?
The reference platform is a Framework Desktop with AMD Ryzen AI Max+ 395 (Strix Halo), 128GB unified memory, and AMD Radeon 8060S iGPU (gfx1151). The unified memory architecture means the iGPU accesses all 128GB directly — 46GB+ LLMs run without a discrete GPU card.
OtherU Core will run on any Linux system with sufficient GPU/unified memory. NVIDIA CUDA is supported via llama.cpp; AMD ROCm requires gfx10xx or gfx11xx.
How does Fara control a computer without installing software on it?
Fara uses a JetKVM device that captures the target computer's HDMI output and injects USB HID (keyboard and mouse) commands. OtherU captures screenshots via WebRTC, runs them through the Fara-7B vision model, and sends HID commands back through the bridge. The target machine sees only a standard USB keyboard and mouse — no drivers, agents, or OS access required.
What LLMs does OtherU use?
| Model | Role | Size |
|---|---|---|
| Qwen3-Next-80B-A3B (MoE) | Primary responses | \~46GB Q4_K_M |
| Nemotron-Orchestrator-8B | Tool routing | \~6.3GB Q6_K |
| DeepSeek-R1-Distill-Qwen-14B | Deep reasoning | \~8.4GB Q4_K_M |
| Fara-7B (microsoft) | Desktop vision | \~16GB |
| Qwen3-Coder-Next | Code generation | \~80 GB UD-Q8_K_XL |
All models run locally via llama.cpp (GGUF) or vLLM (HuggingFace). No external API required.
How does OtherU communicate with its operator?
Via OpenClaw, which connects OtherU Core to Telegram, Signal, and WhatsApp. OpenClaw handles channel routing, scheduling, memory indexing, and skill execution. OtherU Core provides the reasoning and action layer.
What is the GSD loop?
GSD ("Get Shit Done") is an autonomous execution service that implements coding tasks without human intervention. Given an objective, it runs a plan → code → verify loop, self-heals from errors, and reports progress. It targets the OtherU Core codebase, iteratively improving the system.
How is memory handled across sessions?
OtherU uses multiple memory layers:
- Short-term: Redis state and in-session context
- Medium-term: Daily log files (rolling context)
- Long-term: Curated memory file + LightRAG + GraphRAG knowledge graph + LEANN vector index
- Search: OpenClaw indexes all sessions and memory files for semantic retrieval
What is latent-mas?
A research project exploring multi-agent communication via continuous hidden states (latent space) instead of text tokens. Agents "think" in latent space using Coconut/SIM-CoT continuous reasoning, with diffusion-refined thoughts and a consensus mechanism. Early results: GSM8K 38%, MultiArith 88% — research stage.
Can OtherU run without a JetKVM?
Yes. The JetKVM and hardware bridge are only required for Fara desktop automation. All other capabilities — reasoning, coding, memory, messaging, GSD — work without any KVM hardware.
How do I connect OpenClaw to OtherU Core?
Add OtherU Core as a custom provider in OpenClaw:
Type: openai-compatible
URL: http://127.0.0.1:8080/v1
Key: local-dev-key
Model: otheru-core
See the API and Integrations doc for the full endpoint reference.