Deployment platforms
How AI platforms support model access, serving, governance, monitoring, release management, and operations.
Model platforms
Model platforms are where AI access becomes operational infrastructure.
AI model platforms help organizations serve models, route requests, control releases, monitor usage, manage versions, and decide which model should handle which task. They are the technical layer that connects AI applications to governed, observable, and replaceable model access.
What this section explains
These guides cover the platform layer around models: deployment platforms, serving endpoints, gateways, routing, catalogues, versioning, release controls, and rollback paths.
Model serving
How applications call models through endpoints, APIs, runtimes, queues, scaling layers, and response formats.
Gateways and routing
How a gateway can route requests between models, vendors, environments, policies, and fallback paths.
Model catalogues
How organizations keep track of approved models, owners, intended uses, limitations, and status.
Versioning and rollback
How releases, versions, approvals, tests, and rollback controls reduce disruption when models change.
Model platforms article list
This section contains five launch articles. Build these before treating the section as complete.
Start here
Reading order
AI Deployment Platforms Explained
Learn how AI platforms support model access, endpoints, runtime choices, monitoring, permissions, release controls, and production operations.
Serving layerModel Serving Explained
Understand how applications send requests to models and receive responses through serving endpoints and managed runtimes.
Routing layerAI Gateways and Model Routing
See how gateways can route requests, enforce policies, centralize logs, manage fallback, and reduce one-off model connections.
InventoryModel Catalogues and Registries
Learn how model inventories can record ownership, intended use, approval status, versions, risk notes, and retirement status.
Change controlVersioning, Rollback, and Release Controls
Understand why model changes need testing, approval, staged release, fallback plans, and rollback paths.
Recommended path
Start with deployment platforms, then model serving, gateways, model catalogues, and release controls. That follows the path from access layer to operational change management.
How model platforms fit into AI integration
Model platforms sit between AI applications and the models they use. They help keep model access more consistent, observable, controlled, and changeable.
1
Application
A website, internal tool, workflow, agent, dashboard, or support system needs AI output.
2
Platform layer
The platform or gateway applies routing, policy, credentials, logging, and release controls.
3
Model access
The request reaches an approved model, model version, hosted endpoint, or vendor service.
4
Monitoring and change
Usage, errors, latency, cost, quality signals, versions, and rollback paths are tracked.
Operational reminder: The model is not the only thing that matters. The serving layer,
routing rules, release process, monitoring, and fallback plan shape the real integration.
Why model platforms matter
Many early AI projects begin with a direct API call to a model. That may be enough for testing, but production integration usually needs more structure. Teams need to know which model is being used, which version is active, who approved it, how requests are logged, what happens when the model fails, and how changes are rolled back.
A model platform can help manage those concerns. The exact tools vary by organization, but the underlying integration questions are consistent: how model requests are served, routed, monitored, governed, upgraded, and retired.
| Platform concern | Plain meaning | Why it matters |
|---|---|---|
| Serving | How applications call a model and receive output. | Shapes latency, reliability, scaling, and request format. |
| Routing | How requests are assigned to models, vendors, endpoints, or versions. | Supports cost control, quality, fallback, and policy enforcement. |
| Catalogue | A record of approved models, owners, uses, risks, and status. | Prevents unknown or retired models from becoming hidden dependencies. |
| Versioning | Tracking which model version, prompt version, or configuration was used. | Helps explain behaviour changes and supports rollback. |
| Release control | Testing, approval, staged rollout, monitoring, and rollback before major changes. | Reduces disruption when model behaviour changes. |
Platform questions before production use
- Which models or model services are approved for this use case?
- Which applications, users, or workflows can call them?
- Where are requests logged?
- How are latency, errors, cost, and usage monitored?
- Can requests be routed to a fallback model or safe manual path?
- Who approves new models or model versions?
- How are model changes tested before release?
- How can a model, endpoint, prompt, or configuration be rolled back?
How this section connects to the rest of the site
Model platforms depend on the rest of the integration stack. Data systems provide context. APIs and connectors move requests. Identity rules control access. RAG systems retrieve knowledge. Monitoring systems watch behaviour. Security and compliance teams need evidence.
APIs and Connectors
Model platforms usually expose APIs and may route requests through connectors or gateways.
Identity and Access
Model access needs credentials, role boundaries, service accounts, and approval rules.
Monitoring and Observability
Production model platforms need logs, traces, metrics, drift signals, and incident paths.
Security and Compliance
Security review depends on model inventory, access controls, logging, and vendor-risk review.
Educational limitation
This section provides general educational information about model platforms and AI integration. It is not legal, financial, medical, engineering, safety, cybersecurity, procurement, compliance, privacy, or professional advice. It does not provide instructions for bypassing controls, exploiting systems, unauthorized access, or unsafe automation. Use qualified review before using AI model platforms with sensitive data, regulated systems, production infrastructure, customer records, financial processes, safety systems, or other high-consequence environments.