Sessions

A session is the basic unit of work in XClaw. One session = one task = one complete Agent interaction history.

Unlike most AI chat tools, XClaw sessions are designed as parallel, async tasks. You don't need to wait for one to finish before starting the next. The session list on the left works like an inbox, showing each task's status — Todo, In Progress, Needs Review, Done.

Each session has its own context window. Files the Agent has read and operations it has performed in one session don't leak into another. This means you can safely have multiple Agents working on different code branches simultaneously.

Session data is stored locally on disk in JSONL format — an append-only format naturally suited to conversation data that only grows. Your conversations never leave your machine unless you're using a cloud model, in which case messages are sent to the model provider for inference.

When conversations get long, XClaw automatically compacts earlier messages, preserving key context while managing token consumption. This process is transparent to you.

How to

Click the "+" button at the top of the left sidebar to create a new session. You can also use the shortcut Cmd/Ctrl+N for quick creation.

The session list supports filtering and sorting. If you've accumulated many sessions, filter by status (e.g., show only "In Progress") or use the search box to quickly locate one by keyword.

Want to rename a session? Double-click the session title to edit it. Descriptive names like "Refactor Login Module" make it much easier to find things later.

Sessions have independent contexts, but you can use @mention in a new session to reference files from a previous one. If a conversation gets really long, don't worry — XClaw automatically compresses earlier messages to keep the Agent's response quality high.