documentation · memory
Memory model
Memory is what separates a workspace from a chat window. Skyber keeps a store of durable facts attached to your account rather than to a single conversation, so context carries across sessions, projects and models.
Two layers of context
Conversation context
The recent turns of the current thread, plus that conversation's persona and system prompt. Scoped to the thread and bounded so long chats stay responsive.
Account memory
Durable facts about you and your work — stack, conventions, projects, preferences, decisions. Available to every conversation on the account.
How memories are used
- Capture. You can add a memory explicitly, and the assistant can record something worth keeping during a conversation.
- Injection. When you send a message, stored memories are assembled into the system context alongside the persona, so the model starts the turn already knowing your situation.
- Portability. Because memory lives in the workspace rather than the model, switching model or endpoint mid-project keeps every fact intact.
Control and deletion
The memory panel in the product lists everything stored against your account. You can edit an entry, delete one, or clear the lot. Memories are scoped to your account by per-user access rules in the database, and deleting your account removes them along with your conversations and files. See the security overview for details.