Setup and Deployment

System Requirements

OtherU Core runs on a single Linux host with sufficient unified or discrete GPU memory to hold your chosen model stack.

Recommended:

• AMD Ryzen AI Max+ (Strix Halo) with 128GB unified memory, or equivalent
• Linux with ROCm (AMD GPU) or CUDA (NVIDIA GPU)
• Docker + Docker Compose
• 200GB+ storage for models

Minimum (reduced model set):

• 32GB+ GPU/unified memory
• Linux with Docker + Compose

Repository Layout

otheru-core/
├── core/coordinator/          # FastAPI orchestrator (bind-mounted into container)
│   ├── routes/openai_api.py   # OpenAI-compatible API handler
│   ├── council.py             # Multi-tier routing
│   ├── routing.py             # needs_tools() and routing decisions
│   ├── agent_chat.py          # preroute_by_pattern(), orchestrator loop
│   └── tools_impl.py          # All tool implementations
├── core/config/
│   └── routing_policy.json    # Routing weights, tool intent keywords
├── services/hardware_bridge/
│   └── bridge.py              # JetKVM WebRTC bridge
└── infrastructure/docker/
    └── docker-compose.production.yml

Models

Place GGUF models in /opt/models/ and HuggingFace models in /opt/models/vllm/.

Deployment

git clone https://git.otheru.ai/otheru/otheru-core
cd otheru-core
docker compose -f infrastructure/docker/docker-compose.production.yml up -d

Service startup order

1. Core dependencies — Redis, LEANN (RAG)
2. Orchestrator — Nemotron-8B (tool routing)
3. Responder — 80B MoE model (primary responses)
4. Reasoner — 14B R1 model (deep analysis)
5. Coordinator — FastAPI gateway (depends on all agents)
6. Hardware Bridge — JetKVM WebRTC bridge (requires physical JetKVM)
7. GSD — Autonomous dev loop (optional)

Validation Checklist

• [ ] curl http://127.0.0.1:8080/health → {"status":"ok"}
• [ ] curl http://127.0.0.1:8080/dependencies/health → all services healthy
• [ ] curl http://127.0.0.1:8005/stats → frame_age < 100ms (JetKVM live)
• [ ] docker logs otheru-coordinator shows models loaded and routing ready

OpenClaw Integration

OtherU Core exposes an OpenAI-compatible API. Connect OpenClaw as a custom provider:

• Base URL: http://127.0.0.1:8080/v1
• API Key: local-dev-key (or configure your own)
• Models: otheru-core, otheru-council, otheru-coder, otheru-researcher

OpenClaw handles Telegram, Signal, and WhatsApp channels out of the box.

Patching the Coordinator

Coordinator source files are bind-mounted from the host — changes take effect on restart:

# Edit on host, then:
docker restart otheru-coordinator

The hardware bridge is baked into its image. Patch it via docker cp:

docker cp services/hardware_bridge/bridge.py hardware-bridge:/app/bridge.py
docker restart hardware-bridge

WMMA Ops Build (AMD gfx1151)

# Container build (recommended)
docker compose -f docker/docker-compose.benchmark.yml build
docker compose -f docker/docker-compose.benchmark.yml run --rm benchmark bash -c \
  "cd /workspace/wmma_ops && ./build_and_test.sh"

# Host build
pip install -e . --no-build-isolation

Validation gates:

• Correctness checks across representative matrix sizes
• Benchmark pass for adaptive and zerocopy kernel variants (target: 21.6 TFLOPS on gfx1151)
• Regression check against baseline PyTorch matmul