Concept parity — core cloud patterns mapped

This guide maps universal cloud infrastructure patterns—regardless of vendor-specific product branding—directly to Crusoe AI Platform services. Use this concept reference to understand how core architectural building blocks operate on our platform.

Core Cloud Concepts Matrix

Infrastructure Concept	Operational Definition	Crusoe AI Platform Service	Documentation
Virtual Machines	On-demand compute instances running custom software	Deploy containers natively as Serverless Services without managing host operating systems	Deploy a Service
Autoscaling Groups	Automated capacity scaling based on incoming request load	Built-in Autoscaling on all workloads—idle workloads pause and wake on request	Autoscaling & Scale-to-Zero
Load Balancer	Traffic distributor providing public entrypoints and SSL	Every workload has a canonical address, https://<name>-<project-short>.apps.codyhill.dev, that is not reachable from the public internet until you publish it through the gateway; publishing also handles the certificate	Publish an Endpoint
API Gateway	Managed front door providing rate limiting and authentication	Gateway endpoints for publishing, securing, and rate-limiting APIs	Publish an Endpoint
Serverless Functions	On-demand event-driven code execution	Functions: handle(event) entrypoints exposed via HTTPS	Function Quickstart
Object Storage	S3-compatible cloud bucket storage for unstructured files	S3-compatible buckets provided via Crusoe Cloud Integration	Object Storage
Storage Event Triggers	Code run over new objects, delivered on a poll	objectstore Triggers POST the object's own bytes as the request body — not an event record. Non-JSON content reaches a Python handler as {"data": "<text>"}; CloudEvent attributes are exposed under event["_cloudevent"] and on this source do not carry the object key. Delivery is by polling (poll_seconds, default 60), so it is not instant	ObjectStore Triggers
Queue Service	Asynchronous message queuing between producers and consumers	Pub/Sub topics and subscriptions supporting point-to-point queue processing	Pub/Sub Quickstart
Notification Fan-out	One-to-many messaging broadcasting events to subscribers	Pub/Sub Topics with independent subscriber queues tracking positions	Topics & Subscriptions
Event Routing	Event filtering and target delivery based on message rules	Pub/Sub Messaging and workload Triggers	Triggers
NoSQL Key-Value Store	Low-latency in-memory data store for state and caching	MemoryStore: fully managed key-value store that speaks the Redis wire protocol, so any Redis client works	MemoryStore Quickstart
Vector Search	Semantic similarity search over high-dimensional embeddings	VectorDB: dedicated HNSW vector indexing with payload filtering	VectorDB Quickstart
Secrets Service	Encrypted, write-only credentials management	Secrets Manager: write-only secrets injected directly as environment variables	Manage Secrets
Identity & Access Management	Role-based permission policies for users and workloads	Project-scoped admin and member roles plus Service Accounts for machines	Projects & Access
Audit Trail	Immutable security audit logging of management operations	Per-project Audit Log enabled by default for all member actions	Audit Logs
Log Search & Metrics	Metrics collection and log aggregation	Real-time metric counters and 14-day persistent Logs streaming	Troubleshooting
Network Isolation	Isolated network boundaries between tenant workloads	Project-level network isolation blocking unauthorized cross-project access	Workload Networking
Encryption / Key Management	Platform-wide data encryption in transit and at rest	Automated AES-256 platform encryption at rest and TLS 1.3 encryption in transit	Data Protection
Workflow Orchestration	Coordinating multi-step distributed tasks across services	Event-driven Pub/Sub message chains, AI agent tool pipelines, or external webhooks	Agent Tools and Pub/Sub
Container Registry	Image storage and management for containerized workloads	Integrated container image storage managed per workload upon deployment	Crusoe Cloud Integration
Infrastructure as Code	Repeatable declarative resource management	Declarative execution via platformctl CLI scripts and REST API endpoints	CLI Reference
Machine CI Identity	Scoped machine credentials for automated pipelines	Service Account API keys with instant revocation	CI/CD Tutorial

Multi-Step Orchestration Patterns

Building multi-step distributed applications on Crusoe AI Platform is straightforward. Four primary architectural patterns handle complex multi-step workflows:

1. Pub/Sub Queue Chains: Break multi-step jobs into independent processing steps. Each step executes its work, then publishes a message to a Pub/Sub topic triggering the next step. If a worker fails, Pub/Sub redelivers the message automatically.
2. AI Agent Tool Pipelines: Use an Agent to orchestrate multi-step tasks dynamically. The agent uses native Tools to invoke external services and manages execution state via built-in Sessions.
3. Stateful Checkpointing: Store step progress in MemoryStore after each task. If a process restarts, the next step reads the latest checkpoint from MemoryStore and continues processing.
4. External Orchestrator Webhooks: Trigger platform functions or services via HTTPS webhooks from external workflow engines when integrating with existing enterprise control planes.

Advantages of Unified Agent & Serverless Architecture

Traditional clouds force developers to stitch together separate services, configure API gateways, manage IAM policy documents, and coordinate multiple billing line items.

Crusoe AI Platform simplifies this by unifying AI Agents, Serverless Compute, Pub/Sub Messaging, VectorDB, and MemoryStore into a single cohesive project environment:

• One-Command Publishing: Workloads are private when they land; putting one on the internet is a single command with an explicit auth choice (platformctl gateway publish function/hello-http --auth none), with no proxy to run and no certificate to request.
• Built-In AI Agent State: AI agents include automatic session history and long-term memory out of the box—no separate database configuration required.
• Single Developer Toolchain: Manage all compute, agent, messaging, and storage resources using a single CLI (platformctl) and unified web console.
• Predictable Resource Management: Capacity is bounded by real-time project quotas, providing transparent operational visibility across all workloads.

Next Steps

• Read the detailed translation guides: For AWS Users, For GCP Users, and For Azure Users.
• Explore the comprehensive Service Mapping matrix.
• Learn platform fundamentals in Core Concepts.