Skip to main content
An agent has four places to put things: its own database, the shared database, its own folder and the shared folder.

​
Databases

Structured data goes in a database. It is plain SQLite, and it is the default home for anything an agent produces repeatedly: leads, invoices, articles, events, anything with rows.

The agent database

Private to one agent. No other agent can read it.

The shared database

Shared by every agent of the workspace, whichever Box they sit in. This is how a group of agents works from the same tables.
Browse either from the Database tab: the agent panel for the private one, the Knowledge base dialog of the workspace menu for the shared one. You get the table list, the rows with paging, sorting and search, and a Export button for CSV or JSON. You can also delete selected rows or drop a table from there.

​
Etiquette in the shared database

Every agent of the workspace sees the same tables, so a few rules keep it usable:
  • Create with CREATE TABLE IF NOT EXISTS, extend with ALTER TABLE ADD COLUMN.
  • Never drop or alter a table you did not create, and never delete rows you did not insert.
  • Name a table after the domain (leads, invoices), not after the agent that made it.
  • A table has one home. Do not keep a copy in both databases.
The schema view shows which agent created each table, so ownership is never a guess.

​
Files

Files are for deliverables and scratch: a report, a chart, a PDF, an export, a page, some code. Anything with rows belongs in a database instead. Browse and upload from the Files tab of the agent panel, or the Knowledge base dialog of the workspace menu for the shared folder. Dropping a file into an agent workspace is how you give it something to work on.

​
Publishing a file

An agent can publish any file from its workspace to a public URL. The link looks like https://app.rerun.build/p/<slug> and is served to anyone who has it. Re-publishing the same file keeps the same URL, and unpublishing revokes it for real. Limits worth knowing:
  • 25 MB per file.
  • The file is served exactly as it is. Linked assets are not published, so an HTML page has to inline its CSS and its JavaScript.
Ask the agent to publish and it gives you the link. That is usually better than telling you a file is somewhere in its workspace.

​
Getting data out

Export a table

CSV or JSON, from the Database tab.

Query it from the API

db_query and space_db_query return rows straight to your script.

​
Memory is not storage

Memory holds facts about you and about how the work should be done, not the work itself. A preference, a name, a correction: memory. Two thousand rows of leads: the database.