Agent outputs — canvases & the agent database

When an agent runs, its work persists beyond that single execution in two forms. A canvas is something the agent produced — a rendered artifact you can open and read. The agent database is what the agent remembers — a structured store it builds up across runs. Both appear in an agent’s Outputs, and neither is created by hand; they’re products of the agent doing its work.

Canvases

A canvas is a rendered document an agent emits during a run: markdown with embedded charts, tables, and diagrams. You open it from the agent’s Outputs to read the result — a summary, a report, an analysis — as a finished page rather than a chat transcript.

Canvases are read-only views of what the agent produced. If a canvas was created by an automation, you can re-run the agent from the canvas to regenerate it against current data — handy for a recurring report you want refreshed.

Manage: Agents

The agent database

The agent database is an agent’s own structured store. Across runs, an agent can record what it found into tables and read them back next time, so a recurring automation builds on previous work instead of starting cold each time.

Regnora shows this database as a read-only preview: the tables the agent has created, their columns, and the first rows of each. You don’t edit it — it’s a window into what the agent is keeping track of, so you can see the data behind its conclusions.

Output versus memory

It’s worth separating three related ideas:

• A canvas is what an agent produced on a given run — an artifact to read.
• The agent database is structured data one agent accumulates across its own runs.
• Memories are shared know-how in the Library that agents recall between conversations, reviewed and accepted by your team.

Canvases and the database belong to an individual agent and its runs; memories are an organisation-level building block. Reach for the database when you care about the data an agent is tracking, and a canvas when you care about the report it produced.