Files
2dph/compose.yaml
eSliderandGitHub b317968a4c
Tests / Test (push) Skipped
Tests / Release (semver) (push) Skipped
feat: CPU reasoner bake-off for Qwen3.5-9B vs Bonsai (#25)
Measure OpenAI tool_calls (search/get/audit) on a CPU Ollama sidecar
instead of gating D18 on GPU or PicoClaw. Weights stay out of the image.
2026-08-13 23:19:51 +01:00

138 lines
3.7 KiB
YAML

# 2dph — docker composition
#
# docker compose up -d brain # API (Zig CGO serve)
# docker compose --profile index run --rm index # Python rebuild
# docker compose --profile picoclaw up brain-mcp
# docker compose --profile reasoner up -d reasoner # CPU Ollama :11435
# docker compose --profile searxng up -d
#
# Secrets never baked in: search.env + db-profiles.yml from ~/.config/brain.
name: 2dph
services:
brain:
image: ghcr.io/eslider/2dph:api
build:
context: .
dockerfile: Dockerfile
target: api
cache_from:
- ghcr.io/eslider/2dph:cache
command: ["serve"]
environment: &env
HF_HOME: /data/hf
BRAIN_SEARCH_CACHE: /data/cache/web-search.sqlite
BRAIN_DB_PROFILES: /secret/db-profiles.yml
BRAIN_SEARCH_ENV: /secret/search.env
KB_ROOT: /data
KB_WORKERS: "4"
KB_PORT: "8630"
volumes:
- kb-model:/data/hf
- kb-var:/data
- ~/.config/brain:/secret:ro
ports:
- "127.0.0.1:8630:8630"
read_only: true
tmpfs:
- /tmp
healthcheck:
test: ["CMD", "wget", "-qO-", "http://127.0.0.1:8630/health"]
interval: 30s
timeout: 5s
retries: 3
restart: unless-stopped
stop_grace_period: 20s
brain-watch:
image: ghcr.io/eslider/2dph:api
environment: *env
volumes:
- kb-model:/data/hf
- kb-var:/data
- ..:/corpus:ro
- ~/.config/brain:/secret:ro
command: ["watch", "/corpus"]
read_only: true
tmpfs:
- /tmp
restart: unless-stopped
stop_grace_period: 20s
# Python write path (Ladybug rebuild). Not in the API image.
# docker compose --profile index run --rm index
index:
profiles: ["index"]
image: ghcr.io/eslider/2dph:index
build:
context: .
dockerfile: Dockerfile
target: index
environment:
HF_HOME: /data/hf
KB_PY: python3
volumes:
- kb-model:/data/hf
- kb-var:/app/var
- ..:/corpus:ro
- ~/.config/brain:/secret:ro
command: ["index"]
read_only: true
tmpfs:
- /tmp
# Optional local SearXNG (D3). Skip if BRAIN_SEARCH_URL already points at a
# live instance — do not run a second copy on that host.
# SEARXNG_SECRET=$(openssl rand -hex 32) docker compose --profile searxng up -d
searxng:
profiles: ["searxng"]
image: docker.io/searxng/searxng:2026.8.10-0a118066d
ports:
- "127.0.0.1:8888:8080"
environment:
SEARXNG_SECRET: ${SEARXNG_SECRET:-}
volumes:
- ./deploy/searxng/settings.yml:/etc/searxng/settings.yml:ro
- ./deploy/searxng/limiter.toml:/etc/searxng/limiter.toml:ro
restart: unless-stopped
# MCP endpoint for an external agent (PicoClaw is not shipped here).
# docker compose --profile picoclaw up brain-mcp
brain-mcp:
profiles: ["picoclaw"]
image: ghcr.io/eslider/2dph:api
environment: *env
volumes:
- kb-model:/data/hf
- kb-var:/data
- ~/.config/brain:/secret:ro
command: ["serve"]
ports:
- "127.0.0.1:8630:8630"
read_only: true
tmpfs:
- /tmp
restart: unless-stopped
# CPU OpenAI-compatible sidecar (D18). Weights are pulled at runtime, not
# baked into the 2dph image. Does not touch host Ollama on :11434.
# docker compose --profile reasoner up -d reasoner
# docker compose --profile reasoner exec reasoner ollama pull qwen3.5:9b
reasoner:
profiles: ["reasoner"]
image: docker.io/ollama/ollama:latest
environment:
OLLAMA_NUM_GPU: "0"
OLLAMA_HOST: "0.0.0.0:11434"
ports:
- "127.0.0.1:11435:11434"
volumes:
- reasoner-ollama:/root/.ollama
restart: unless-stopped
volumes:
kb-model:
kb-var:
reasoner-ollama: