Appearance
Deployment Guide
GhostAI supports five deployment targets, each with different trade-offs for complexity, cost, and data residency. This guide covers every option with architecture diagrams and complete .env examples.
Deployment Targets at a Glance
| Target | Infrastructure | Database | LLM Options | Best For |
|---|---|---|---|---|
| Native (Mac/Linux) | Local processes + Docker PostgreSQL | PostgreSQL | Ollama, llama-cpp | Solo development, air-gapped demos |
| Docker Compose | Docker containers | PostgreSQL + Redis | Any provider | Team development, on-premise |
| Production Docker | Docker + Caddy (HTTPS) | PostgreSQL + Redis | Any provider | Self-hosted production |
| AWS ECS | ECS Fargate, RDS, ElastiCache | PostgreSQL (RDS) | Cloud providers | Enterprise cloud |
| Kubernetes | Helm charts, K8s | PostgreSQL + Redis | Any provider | Multi-tenant, high scale |
Choosing a Deployment
LLM Provider Architecture
Every deployment uses the same three-layer LLM architecture. Each layer is independently configurable. GhostAI supports a multi-provider architecture with 11 direct providers, OpenRouter, and three local backends (Ollama, vLLM, llama-cpp):
Key insight: You can mix cloud and local. For example, use OpenAI for chat but llama-cpp for PII -- your sensitive data never leaves the machine even though chat uses the cloud.
Next Steps
- Native Deployment -- simplest, zero containers
- Docker Deployment -- recommended for most users
- Production Deployment -- self-hosted with HTTPS
- AWS Deployment -- enterprise cloud
- Kubernetes Deployment -- multi-tenant scale
- Scaling Guide -- horizontal/vertical scaling, RDS connection pool, timeouts
- LLM Providers Deep Dive -- all provider options explained