groot
d18405ed2e
ci / test (services/knowledge) (push) Successful in 47s
ci / test (services/agent) (push) Successful in 55s
ci / test (platform/neuroforge) (push) Successful in 56s
ci / docker-build (push) Failing after 1m10s
ci / test (services/control) (push) Successful in 1m26s
release-tag / release-image (push) Successful in 6m43s
Master Deployment
The Master compose deliberately contains no local CPU/GPU NeuroForge workers. It uses remote subagents. Agent, Knowledge and NeuroForge use OLLAMA_BASE_URL/OLLAMA_URLS, normally pointing to the GPU subagent host.
Before first start:
- replace the
192.0.2.xexample addresses; - fill the five GLPI credentials in
.env; - expose the NeuroForge Master port only to trusted subagent networks or put it behind TLS/mTLS/reverse proxy;
- keep Grafana and Prometheus bound to loopback unless intentionally proxied.
Start:
./preflight.sh
docker compose --profile research --profile monitoring up -d
Monitoring:
- Prometheus:
http://127.0.0.1:${PROMETHEUS_HOST_PORT:-9090} - Grafana:
http://127.0.0.1:${GRAFANA_HOST_PORT:-3000} - Dashboard is provisioned automatically in folder
NeuroForge.