Deploy VibOps on your infrastructure in minutes. This repository contains everything needed to run VibOps — no source code, just configuration and pre-built Docker images.
- Linux server: 4 vCPU, 8 GB RAM, 50 GB SSD minimum
- Docker 24+ and Docker Compose v2
- An LLM API key (Claude recommended) or on-prem LLM (Ollama, vLLM)
No GPU required on the VibOps server — GPUs stay on your clusters.
# 1. Clone this repository
git clone https://github.com/VibOpsai/vibops-install.git
cd vibops-install
# 2. Authenticate to the VibOps container registry (token provided by VibOps)
make login VIBOPS_REGISTRY_TOKEN=<your-token>
# 3. Start the full stack (generates secrets, starts all services)
make quickstart
# 4. Set your LLM provider in .env
# Edit LLM_PROVIDER and LLM_API_KEY, then:
docker compose restart agent
# 5. Create your organisation
make pilot-create-client ORG="My Company" EMAIL=admin@company.com PASSWORD=yourpassword
# 6. Open the console
# http://SERVER_IP:8003| File | Purpose |
|---|---|
docker-compose.yml |
Full stack: core, agent, console, worker, PostgreSQL, Redis, Prometheus, Grafana |
.env.example |
Environment template — all variables documented |
Makefile |
Quickstart, health checks, client provisioning, backups |
prometheus.yml |
Prometheus scrape configuration |
alerting_rules.yml |
GPU and service alerting rules |
grafana/ |
Grafana datasources and pre-built dashboards |
scripts/ |
Health check, gateway setup, onboarding automation |
docs/installation.md |
Complete installation guide (13 sections) |
| Service | Port | Description |
|---|---|---|
| Console | 8003 | Web UI — open in browser |
| Core API | 8000 | REST API + job engine |
| Agent | 8001 | LLM agent |
| LLM Proxy | 8004 | OpenAI-compatible inference proxy — per-agent GPU cost attribution |
| Grafana | 3000 | Dashboards (admin / auto-generated password) |
| Prometheus | 9090 | Metrics |
The full interactive API documentation is available at:
- Swagger UI:
http://SERVER_IP:8000/docs - ReDoc:
http://SERVER_IP:8000/redoc
Both endpoints are enabled by default (APP_ENV=development). In production mode (APP_ENV=production), they are disabled and return 404 for security.
VibOps includes a transparent OpenAI-compatible proxy that tracks GPU cost per AI agent. Point your agents at the proxy instead of your LLM server:
# Your agents call the proxy instead of vLLM/Ollama directly
OPENAI_BASE_URL=http://SERVER_IP:8004/v1
# Add a header to attribute costs per agent
curl -X POST http://SERVER_IP:8004/v1/chat/completions \
-H "X-VibOps-Agent-Id: pricing-agent" \
-H "X-VibOps-Team: supply-chain" \
-d '{"model": "mistral:7b", "messages": [...]}'Per-agent cost attribution is visible in the console under FinOps → Agent LLM Usage.
GPU clusters connect to VibOps via an outbound-only gateway (no inbound ports required on the cluster side):
# From the VibOps console: Fleet → Add Gateway → copy the token
# On the GPU cluster:
helm upgrade --install vibops-connect vibops/vibops-connect \
--set vibops.coreUrl="https://your-vibops-server" \
--set vibops.token="<token-from-console>"VibOps starts a 14-day trial automatically (10 GPUs / 5 users / 2 clusters). Contact david@vibops.ai for a production licence key.
See docs/installation.md for the complete guide including:
- Helm production deployment
- On-prem LLM (air-gapped / sovereign)
- Team management and RBAC
- Configuration reference
- Upgrade and backup procedures
If something isn't working, generate a debug bundle and send it to VibOps support:
make debug
# → vibops-debug-2026-06-24-143000.tar.gz
# Send this file to david@vibops.aiThe bundle collects logs, container status, health checks, resource usage and configuration — all secrets are automatically redacted.
- Email: david@vibops.ai
- Documentation: https://vibops.ai/docs