Docker (recommended for most deployments)
| Dependency | Version | Notes |
|---|---|---|
| Docker Engine | 20.10+ | Or Docker Desktop |
| Docker Compose | v2+ | For the compose-based installation |
Kubernetes (Helm)
| Dependency | Version | Notes |
|---|---|---|
| Kubernetes | 1.24+ | Any conformant cluster |
| Helm | 3.10+ | For installing the ManyLayers chart |
| kubectl | matching cluster | For inspection and troubleshooting |
Infrastructure services
ManyLayers requires Postgres and optionally benefits from Redis, Kafka, and Qdrant as you scale. The Docker Compose bundle starts Postgres automatically for single-host deployments.PostgreSQL 14+
Required. The compose bundle starts Postgres automatically. For production, point ManyLayers at your existing managed Postgres instance.
Redis 7+
Recommended for multi-replica deployments. Shares rate-limit, budget, and cache state across all gateway instances. Without Redis, limits are enforced per-process.
Kafka 2.8+
Optional. Provides durable queuing for high-throughput background work (connector syncs, workflow dispatch, training jobs). The default DB-backed queue works well for single-replica deployments.
Qdrant 1.9+
Recommended for production RAG knowledge bases and semantic caching. An in-memory vector store is available for evaluation and single-replica use.
You do not need GPU hardware to evaluate ManyLayers. The built-in mock LLM provides an OpenAI-compatible endpoint that echoes requests, letting you explore all gateway features before connecting a real AI provider.