Dropstone Support

Agent runtime

An agent runtime is the engine an agent actually runs on: it holds the session, decides which tools to call, streams the reply, and enforces limits. In Dropstone the same runtime serves chat, the CLI, and the SDK.

An agent runtime is the engine an agent runs on. The model is one part of it: the runtime is what holds the session, decides which tools to call and in what order, streams the reply back, keeps memory in view, and enforces what the account is allowed to do.

It is the layer that makes an agent an agent rather than a chat completion. A completion answers one message. A runtime can read a file, run a command, wait for your approval on an action, call back into memory, and carry a task across many steps.

In Dropstone the same runtime serves every surface. Dropstone Chat and the CLI are two clients of it, and the SDK gives your own code the same one: sessions, tools, streaming, and the account's memory. That is why a session your application starts knows what you taught the CLI.

One consequence worth knowing: the public HTTP API is not the runtime. It is a stateless, OpenAI-compatible completions endpoint, so it carries no memory and no tools. If you want the agent in your code, the SDK's local client is the route.

Related articles

Ctrl+I