Your first memory

Capture something worth remembering, then read it back.

Ocean Labs

With your agent connected, capturing memory is one statement. There’s nothing to groom by hand.

Capture#

Your agent writes what’s worth remembering as it works:

insert into entries (content)
values ('The user prefers concise answers and dislikes hype.');

That’s it. The entry lands in your personal workspace, attributed to you. No workspace_id needed, no author_id to set — both default to you.

A fuller capture adds a title, a type, and a label or two:

insert into entries (title, content, entry_type)
values ('Tone preference', 'Prefers concise answers, dislikes hype.', 'decision')
returning id;

-- then tag it (one call upserts the tags and links them):
select apply_tags('entry', '<that-entry-id>', array['preferences', 'tone']);

See Capturing memory for entry types, projects, and tags.

Retrieve#

Later, the agent reads back what it needs. The raw log is always there:

select content from entries order by created_at desc limit 20;

But the better answer usually lives in the wiki, where the Dreamer has already consolidated and cited things. Read pages first:

select title, summary from pages
where status = 'active'
  and fts @@ websearch_to_tsquery('english', 'tone preference');

Then follow links down to the source entries when you need provenance. See Organizing memory.

The Dreamer#

You don’t organize memory yourself. The Dreamer runs nightly: it consolidates entries into pages, links what relates, and keeps the whole thing tidy — so memory compounds instead of just piling up. See The Dreamer.

Nothing is mined, nothing is sold, and nothing is deleted behind your back.