Deployment & Self-Hosting

DMOps.ai can be deployed in two ways: managed cloud (hosted by Dynamics Mobile on AWS) or self-hosted (you run it in your own infrastructure).

---

Managed cloud

For most organisations, the managed cloud option is the right choice. Dynamics Mobile handles:

• Infrastructure provisioning and scaling
• Database backups
• Security patching
• LLM provider key management

Contact sales@dynamicsmobile.com to get your managed tenant.

---

Self-hosting with Docker Compose

For development, evaluation, or air-gapped environments, you can run DMOps.ai locally using Docker Compose.

Prerequisites

• Docker Desktop 4.x or Docker Engine 24+
• At least 4 GB RAM available to Docker
• An API key for at least one LLM provider (AWS Bedrock, Anthropic, or OpenRouter)

Quick start

# Clone the repository
git clone https://github.com/MobileAffairsLTD/dm-acp.git
cd dm-acp

# Copy and edit the environment file
cp .env.example .env
# Edit .env — set HERMES_ACP_API_KEY, HERMES_ACP_MASTER_KEY, and your LLM provider key

# Start the dev stack
bash deployments/docker/deploy.sh --dev

The UI will be available at http://localhost:5173. The API will be available at http://localhost:8001.

Environment variables

Variable	Required	Description
HERMES_ACP_API_KEY	✅	API key for the DMOps REST API
HERMES_ACP_MASTER_KEY	✅	Master encryption key for integration credentials
PG_ENABLED	—	Set to 1 to use PostgreSQL instead of SQLite
DATABASE_URL	If PG_ENABLED	PostgreSQL connection string
ACP_MAX_CONCURRENT	—	Max concurrent agent tasks (default: 4)

---

Self-hosting on AWS Fargate (production)

For production deployments, DMOps.ai runs on AWS ECS Fargate with a dedicated PostgreSQL database per tenant on Amazon Aurora Serverless.

Architecture

Internet → ALB → ECS Fargate (DMOps API)
                       ↓
                 Aurora Serverless (PostgreSQL)
                       ↓
                 S3 (artefact storage)

Sizing

Tier	vCPU	Memory	Max concurrent tasks	Typical use
Entry (T1)	0.5	1 GB	2	Dev/demo, 1–3 agents
Light (T2)	0.5	2 GB	4	3–6 agents
Standard (T3)	1.0	4 GB	8	6–12 agents
Growth (T4)	2.0	8 GB	16	12–24 agents
Scale (T5)	4.0	16 GB	32	24+ agents

Deploying

Use the acpfleet.py CLI to provision and manage tenants:

# Provision a new tenant
python acpfleet.py add <tenant_id> --tier 2

# Scale up
python acpfleet.py update <tenant_id> --tier 3

# Roll out a new image (zero downtime)
./deployments/rollout-image.ps1 -ImageTag <tag>

Never use terraform apply for image-only updates

Rolling out a new image must be done via rollout-image.ps1. Running acpfleet.py update for image changes triggers an ECS service replacement (destroy + recreate), causing ~5 min of downtime per tenant.

---

Health checks

The API exposes two health endpoints:

Endpoint	Purpose
GET /api/ping	Liveness check — returns {"status": "ok"}
GET /api/stats	Readiness check — includes DB connectivity and dispatcher status

---

Backups

On self-hosted PostgreSQL, configure your own backup policy. On managed Aurora Serverless, automated backups are enabled with a 7-day retention window.

Artefact files stored in S3 should have versioning enabled and a lifecycle policy for long-term retention.