Interactive CLI vs Agent SDK
Also called: interactive session, programmatic entry point, print mode
What is the difference between the interactive CLI and the Agent SDK?
The interactive CLI is the terminal session a developer runs directly, with permission prompts and plan gates. The Agent SDK and non-interactive print modes are programmatic entry points that drive the same model without those gates, and route the work through metered API usage instead of the local subscription.
Two ways to reach the same model.
They look interchangeable from outside and behave differently in the two places that matter most: who supervises, and who pays what.
The interactive CLI is a supervised session
It is the terminal session a developer starts. Consequential actions pause for a permission answer, plan mode gates on approval, and the work runs under the Claude subscription installed on that machine.
Programmatic entry points are built for automation
An SDK, a stream-json transport, or a non-interactive print mode exists so software can drive the model without a person in the loop. The interactive gates are not there, because a caller would have nothing to answer them with.
The billing follows the path, not the work
The same task can be a flat subscription cost through the interactive CLI or metered API usage through a programmatic interface. Nothing about the task changed. The entry point did.
Why anyone should care.
This is normally an implementation detail. It stops being one as soon as agents run unattended.
Predictable cost
A flat subscription does not move when a run takes six hours. Metered usage does, and the number is only knowable afterwards.
Supervision points
Permission prompts and plan gates only exist in the interactive path. Remove them and there is nothing for a human to approve remotely.
Where the credentials live
The interactive CLI uses the subscription on the machine. A programmatic path needs API credentials held by whatever is calling it.
Commonly confused with.
Three distinctions worth keeping straight.
Integration path questions
What is the practical difference between the two?
The interactive CLI is a session a person supervises, running under the Claude subscription on that machine. Programmatic entry points are built for automation: they remove the interactive gates and bill as API usage, so cost scales with tokens rather than sitting flat.
Which one does Virex Remote use?
The interactive CLI, deliberately. That keeps sessions billed under your own subscription and keeps the permission and plan gates that remote approval acts on.
Is the Agent SDK worse?
No, it is built for a different job. If you are embedding an agent inside a product you ship, a programmatic interface is the right tool. If you want to supervise a developer session running on a real machine, the interactive CLI is what has the supervision points in it.
Does the interactive CLI limit what an agent can do?
Not in capability. It differs in that consequential actions pause for an answer. That pause is a feature when a human is accountable for the run, and an obstacle when nobody is meant to be.
Why does this distinction show up in pricing conversations?
Because it decides whether a long unattended run costs a predictable flat seat or a variable amount that depends on what the agent consumed. The integration method is upstream of the invoice.
Oversight without a metered bill.
Virex Remote drives the interactive CLI, so a seat is the whole cost.