A SaaS consumer uses a vendor-hosted AI product, with the provider managing the application, model, and retrieval pipeline. The organization relies on vendor controls and contractual guarantees rather than operating the underlying stack. section 9.3.1
Context engineering in this posture is limited to the controls the vendor exposes:
- Use prompt patterns, user-supplied context, and administrative knowledge grounding. section 9.3.1
- Configure exposed options such as data connectors, retention policies, access controls, and audit logging; do not expect direct control of system instructions, retrieval algorithms, or model parameters. section 9.3.1
Enterprise products
ChatGPT Enterprise, Claude Enterprise, and Microsoft 365 Copilot illustrate vendor-hosted products in this adoption posture. 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 controls, and low operational burden with other adoption models. section 9.3.1

