Saving a memory

Space is in development. These guides describe the current development version and the intended workflow. Public sign-ups and hosted agent connections are not available yet.

Saving memory starts with an explicit action by your agent.

The agent sends something to Space

Connecting an agent gives it access to tools. The agent still has to use a tool to save your preference. Space does not automatically collect your conversations.

There are two ways to create claims

• Direct writing: the agent sends a finished statement, such as “The user prefers Python for scripting.” The technical name for this is structured input.
• Extraction: the agent sends text or points to a saved document. A background processor reads it and proposes statements to remember.

Both paths produce claims. Extraction can take time, so an accepted job is not a finished memory. Once claims are successfully saved and indexed, they are available for retrieval without a human approval step.

Indexing prepares each claim for search

Think of a book’s index: it helps you find a passage. Space creates a search entry for each current claim, with a link back to the saved text and numbers that help compare meaning. The next guide explains those numbers.

What happens when a claim changes?

• New claim: Space creates its search data.
• Changed statement: Space updates the search data to describe the new text.
• Unchanged statement: it stays searchable using the existing data, unless the model used to create that data changes.
• Forgotten claim: its entry is removed from active memory search.

This is why “selected for indexing” can be confusing. All current claims have entries; only some need fresh search data during an update. Being unchanged does not make a claim disappear from search.

Forgetting a claim does not erase its text from source documents, retained revisions, or backups. See Privacy & data for those separate boundaries.