LLM off-the-shelf is an adoption posture in which an organization deploys a provider-supplied or open-weight model in its own or a managed environment. The organization manages serving, model updates, and application integration. Its context engineering scope includes the context layer, serving configuration, adaptation, and context caching; its operational burden is high. section 9.3.1
Table 9.4 Adoption models for context-engineered systems
| Adoption model | Operating boundary | Context engineering control | Operational burden |
SaaS consumer |
Vendor hosts the application and the model |
Prompts, plus the connectors, retention options, access controls, and audit logs the vendor exposes |
Low |
API integrator |
Organization owns the application, provider hosts the model endpoint |
Full context layer: retrieval, tools, memory, state, guardrails, evaluation, and tracing |
Medium |
LLM off-the-shelf |
Organization deploys a provider-supplied or open-weight model in its own or a managed environment |
Context layer, plus serving configuration, adaptation, and context caching |
High |
Model hoster |
Organization operates the model platform in a private network, VPC, or regulated environment |
Context layer, plus model serving, logs, security boundaries, data locality, and model lifecycle |
Very high |
Compare this posture’s operating boundary, context engineering control, and burden with SaaS consumer, API integrator, and model hoster approaches. section 9.3.1
