Agent Deployment Model (Hosted vs. Self-Hosted)
Agent SDKDefinition
The choice of where an agent's loop and tools run: Anthropic-hosted/managed (Anthropic runs the infrastructure, lowering operational burden but reducing environment control) versus self-hosted/in-process (the developer runs the loop and tools in their own environment, maximizing control over data flow, networking, and least privilege at the cost of operating it themselves). Neither option is universally correct -- the decision trades operational burden against control over data residency and least privilege.
Example Usage
A regulated workload with strict data-residency requirements favors self-hosting the agent loop, even though it means operating the infrastructure directly, because a hosted deployment would trade away control over exactly where and how data flows.
In Depth
The Two Models
| Model | Character |
|---|---|
| Anthropic-hosted / managed | Anthropic runs the agent infrastructure; lower operational burden, less control over the environment |
| Self-hosted (in-process) | The developer runs the loop and tools in their own environment; maximum control over data flow, networking, and least-privilege boundaries, at the cost of operating it themselves |
This decision sits alongside, but is distinct from, the choice between the Claude Agent SDK and a hand-rolled custom loop. An SDK-driven loop can be deployed either hosted or self-hosted, and a custom loop is typically self-hosted by construction -- the two choices (SDK vs. custom loop; hosted vs. self-hosted) are orthogonal design decisions.
Why Neither Wins Universally
Hosted deployment removes the burden of running and scaling the agent infrastructure, which is attractive when the team doesn't want to own that operational surface. Self-hosting trades that convenience for tighter control over exactly where data flows, what network boundaries apply, and how tightly tool access is scoped -- control that some data-residency, compliance, or least-privilege requirements demand outright.
Exam Framing
The decision follows the same requirement-driven logic used throughout application design: read the actual constraint (data residency, compliance, operational capacity) rather than assuming the more "convenient" or more "advanced"-sounding option is automatically correct.
Common Pitfalls
- Assuming hosted/managed deployment always wins because it's less operational work -- self-hosting is a legitimate, sometimes required, choice when data residency or least-privilege control is the actual constraint.
- Conflating the hosted-vs-self-hosted decision with the Agent-SDK-vs-custom-loop decision -- they are independent axes, not the same choice described twice.