Session Backends

Session backends are stateful server-side processes that are bound to the life of a user’s usage interaction session with an application. Client-side code can connect directly to a running session backend using a tokenized URL.

Typically, an application will start a session backend in response to some user action (such as opening a document). Session backends are terminated when the user navigates away from that document, or closes the browser tab.

Unlike a traditional web backend server, session backends are not expected to be stateless or interchangeable. Instead, session backends provide a way to guarantee that every request from a particular user (or group of users) is handled by the same process, and no other requests are.

You can use a session backend to, for example, load a single user’s large dataset into memory, and then serve low-latency requests to the user. Or, you could use a session backend as a real-time collaboration server, and have multiple users connect to the same backend to synchronize their changes.

With appropriate protections, session backends can also be a good sandbox for running untrusted user- (or LLM-) provided code, since they can be isolated to a particular user and treated as disposable.

Although the term session backend is relatively new, the concept is not. Figma has written (opens in a new tab) about their multiplayer approach which could be described as a session backend, as could the architecture of the Jupyter (opens in a new tab) scientific computing environment and VS Code for the web (opens in a new tab).

Read more about session backends (opens in a new tab).