Deployment Overview

Agentcy is a self-hostable platform. You can run it anywhere that supports Linux containers or a static binary. Pick the option that matches your operational comfort, scale, and compliance requirements.

Decision Matrix

Option	Best for	Effort	HA	Managed databases
Local Development	Building features, testing locally	Lowest	No	No
Docker Compose	Single VM, small teams, demos	Low	No	No
Bare Metal	Single Linux host, no Docker, air-gapped sites	Low–Medium	No	Optional
Desktop App	Individual users on macOS	Lowest	No	No
Railway	Fastest cloud path, hobby–small team	Low	Limited	Yes (PG, Redis)
AWS CloudFormation	Single-account AWS, push-button stack	Medium	Yes	Yes (RDS, ElastiCache)
AWS ECS	AWS shops on Fargate, no k8s	Medium	Yes	Yes
AWS EKS	AWS shops standardized on Kubernetes	High	Yes	Yes
Kubernetes (any)	GKE / AKS / on-prem k8s, GitOps shops	High	Yes	Optional
GCP	Cloud Run for stateless tier, GKE for full HA	Medium–High	Yes	Yes

Architecture Recap

Every deployment runs the same components:

Component	Process	Default port	Notes
agentcy-api	Rust/Axum backend	8080	Single binary, scales horizontally
agentcy-frontend	Next.js 16 static + server	3000	Stateless, scales horizontally
PostgreSQL	Managed or self-hosted	5432	Source of truth — relational data
Context Engine	Embedded graph store (Basic) OR kyma sidecar (Advanced)	7687 (Basic, Bolt) / 8080 (Advanced, HTTP + Flight gRPC)	Knowledge graph — see Architecture & Tech Stack
Redis	Managed or self-hosted	6379	Job queues, caching
OpenFang (optional)	Sidecar	4200	Sub-agents and orchestration

The backend runs database migrations on startup. There is no separate migration step.

Sizing Guidance

Starting points — adjust based on your workload (tools/min, agent concurrency, graph size):

Tier	Users	API	Frontend	PostgreSQL	Neo4j	Redis
Small	1–10	1 × 0.5 vCPU / 512 MB	1 × 0.25 vCPU / 256 MB	1 vCPU / 1 GB	1 vCPU / 2 GB	0.25 vCPU / 256 MB
Team	10–50	2 × 1 vCPU / 1 GB	2 × 0.5 vCPU / 512 MB	2 vCPU / 4 GB	2 vCPU / 4 GB	0.5 vCPU / 512 MB
Org	50–500	3+ × 2 vCPU / 2 GB	2 × 1 vCPU / 1 GB	4 vCPU / 16 GB	4 vCPU / 16 GB	1 vCPU / 1 GB

The Rust backend is memory-light. Most workloads stay well under 1 GB per replica.

Choosing the Context Engine

Agentcy ships two interchangeable Context Engine providers behind one agentcy-graph interface — Cypher works on both, application code is provider-agnostic.

Provider	Backed by	Query languages	Where it runs
Basic (default)	Embedded graph store, Neo4j-compatible	Cypher	Bundled with every self-hosted install. Optionally swap in Neo4j AuraDB or self-managed Neo4j Enterprise for HA.
Advanced	kyma — Rust columnar engine on S3 + pgvector	KQL / SQL / Cypher	Agentcy Cloud, on-premise Enterprise, and self-host for evaluation. Set CONTEXT_ENGINE=advanced and point at a kyma sidecar.

The Advanced provider unlocks structured KQL/SQL, time-series queries, schema evolution, multi-billion-node scale, and Apache Arrow Flight streaming. On Agentcy Cloud the engine, S3 bucket, and pgvector are provisioned for you — bring-your-own-bucket (AWS S3, GCS, Cloudflare R2, MinIO) is supported for data residency. See Architecture & Tech Stack → Context Engine for the full breakdown and the kyma project at getkyma.dev.

Production Checklist

Before going live on any platform:

• [ ] All default passwords replaced (POSTGRES_PASSWORD, JWT_SECRET, plus context-engine creds — NEO4J_PASSWORD on Basic or KYMA_TOKEN on Advanced)
• [ ] TLS terminated in front of the API and frontend
• [ ] CORS_ORIGINS restricted to your real domains
• [ ] Database backups scheduled (PG: pg_dump nightly; Neo4j: dump before upgrades)
• [ ] Logs shipped off the host (CloudWatch, GCP Logging, Loki, Datadog)
• [ ] Metrics scraped (/metrics Prometheus endpoint on the API)
• [ ] Auth provider chosen — local for solo, oidc for teams (see OIDC)
• [ ] Resource limits set on every container/pod
• [ ] Zero-trust policies enabled if needed (see Policies)

Getting Help

• Self-host support: open an issue at github.com/agentcylabs/agentcy (release tarball customers also get email support)
• Managed offering: Agentcy Cloud takes deployment off your plate entirely
• Custom environments (air-gapped, FedRAMP, on-prem k8s): contact agentcylabs.com