Skip to Content
DeploymentOverview

Deployment

M3 Forge now has a deployment story built around hosting models, not just container entrypoints.

The target platform family is:

  • Cloud: Marie AI operates both control plane and runtime
  • Hybrid: Marie AI operates the control plane, you operate the runtime
  • Self-hosted: you operate both
  • Legacy local packaging: Docker-based demo and migration paths

The production-oriented path is Kubernetes + Helm. The Docker pages remain useful for demos, local environments, and migration, but they are no longer the architecture target for serious self-hosting.

Hosting Models

Cloud Provider Guides

Current State

CapabilityStatus
Standalone m3forge Helm chartAvailable
API + frontend split workloadsAvailable
PVC-backed control-plane filesystem pathsAvailable
Ingress and Gateway API chart supportAvailable
Full umbrella chart with marie-aiIn progress
Hybrid listener/operator packageIn progress
Kubernetes-native webapp runnerIn progress

Platform Prerequisites

All serious deployment models eventually rely on the same base services:

  • PostgreSQL for transactional data
  • S3-compatible object storage for documents and artifacts
  • ClickHouse for production observability and analytics
  • Kubernetes for the target production deployment model
  • Argo CD (sandbox/snapshot feature only) — GitOps control plane that reconciles each sandbox into a complete, isolated Marie system in its own namespace

Today, M3 Forge expects an S3-compatible object storage API. AWS S3 works directly. On GCP and Azure, use MinIO or another S3-compatible layer until native provider adapters are introduced.

Legacy Local Packaging

The Docker pages are still useful when you need:

  • a quick demo environment
  • a local migration path from Docker to Kubernetes
  • a developer workstation setup

Architecture

Next Steps

Last updated on