ODS V3 Pre-Release: Public testing and refinement ahead of the official V3 launch. Turn your PC, Mac, or Linux box into a private AI server.
Python ai-agentsamdcomfyuidockerllama-cppllmlocal-ain8nnvidiaopen-webuiragself-hostedspeech-to-textstrix-halotext-to-speechworkflow-automation
# ODS -- macOS Apple Silicon Overlay
# llama-server runs natively on macOS host with Metal acceleration
# All other services remain in Docker, reaching the host through the scoped
# Colima vmnet gateway (or host.docker.internal on Docker Desktop).
#
# Use with: docker compose -f docker-compose.base.yml -f installers/macos/docker-compose.macos.yml up -d
services:
llama-server:
image: busybox:1.36.1@sha256:73aaf090f3d85aa34ee199857f03fa3a95c8ede2ffd4cc2cdb5b94e566b11662 # Placeholder -- container never runs (replicas: 0)
deploy:
replicas: 0 # Disabled -- running natively on macOS host via Metal
# Readiness sidecar — polls native llama-server on the macOS host.
# Docker Compose cannot express depends_on for a host process, so this
# lightweight container bridges the gap. open-webui waits on its
# healthcheck before starting.
llama-server-ready:
image: busybox:1.36.1@sha256:73aaf090f3d85aa34ee199857f03fa3a95c8ede2ffd4cc2cdb5b94e566b11662
container_name: ods-llama-ready
restart: unless-stopped
extra_hosts:
- "host.docker.internal:host-gateway"
entrypoint:
- sh
- -c
- |
echo "Waiting for native llama-server on host..."
while ! wget -qO- "http://${ODS_MACOS_HOST_GATEWAY:-host.docker.internal}:${OLLAMA_PORT:-8080}/health" >/dev/null 2>&1; do
sleep 2
done
echo "llama-server is ready"
sleep infinity
healthcheck:
test: ["CMD-SHELL", "wget -qO- http://${ODS_MACOS_HOST_GATEWAY:-host.docker.internal}:${OLLAMA_PORT:-8080}/health || exit 1"]
interval: 10s
timeout: 5s
retries: 60
start_period: 5s
deploy:
resources:
limits:
cpus: '0.1'
memory: 32M
security_opt:
- no-new-privileges:true
dashboard-api:
environment:
- OLLAMA_URL=http://${ODS_MACOS_HOST_GATEWAY:-host.docker.internal}:${OLLAMA_PORT:-8080}
- OLLAMA_HOST=${ODS_MACOS_HOST_GATEWAY:-host.docker.internal}
- OLLAMA_PORT=${OLLAMA_PORT:-8080} # llama-server runs natively on macOS at this port
# gpu.py routes to get_gpu_info_apple() which reads HOST_* env vars.
# config.py defaults include 'apple' so all manifests are discovered.
- GPU_BACKEND=apple
# Static host hardware info written to .env by install-macos.sh (Phase 2).
- HOST_RAM_GB=${HOST_RAM_GB:-0} # explicit override; base.yml also carries this for other platforms
open-webui:
depends_on:
llama-server-ready:
condition: service_healthy
environment:
OPENAI_API_BASE_URL: "${OPEN_WEBUI_LLM_BASE_URL:-http://${ODS_MACOS_HOST_GATEWAY:-host.docker.internal}:${OLLAMA_PORT:-8080}/v1}"
OPENAI_API_KEY: "${OPEN_WEBUI_LLM_API_KEY:-}"