Memory

XClaw's memory system addresses a fundamental problem with Agents: every conversation starts from scratch.

An Agent without memory doesn't know if you prefer TypeScript or JavaScript, doesn't know your project uses a monorepo structure, doesn't know you decided last week to switch state management from Redux to Jotai. Every time, you have to re-explain this context.

The memory system changes that. The Agent automatically extracts things worth remembering during conversations: your technical preferences, recurring architectural decisions, project-specific conventions. These memories are persisted and automatically injected into future session contexts.

You have full control over this process. Review what the Agent remembers in Settings, delete outdated entries, or manually add information you want the Agent to always know. Auto-memory can be turned off entirely.

In team workspaces, memories can be shared across members. This means team coding standards and project conventions — write them once, and every member's Agent picks them up.

Memory data is stored locally, just like all other user data.

How to

Open Settings > Memory to see a list of everything the Agent has automatically remembered. Each entry shows the source session and timestamp.

Think a memory is outdated? Click the delete button next to it. Want to add one manually? Click "Add Memory" and type whatever you want the Agent to always know — like "Our project uses pnpm as the package manager" or "API responses must follow the JSON:API spec."

If you don't want the Agent to auto-remember at all, there's an "Auto Memory" toggle at the top of the page. Turn it off and the Agent won't extract memories from conversations, though manually added entries still work.

In team workspaces, admins can manage shared memories in the workspace settings, ensuring all members' Agents follow the same project conventions.