Three things make an agent more than a one-off chat: what it remembers, what it reads, and what it can reach.
Memory
Each agent carries durable facts into every run — your preferred report format, naming conventions, the context it keeps needing. Memory updates automatically after runs, and you can read or edit it yourself on the Memory tab. Prefer a clean slate each time? Turn memory off in Settings.
Skills
Skills are documents loaded into the agent's context on every run — style guides, checklists, templates, anything it should always have on hand. Add any workspace document as a skill on the Skills tab. Presets arrive with their own skills, kept in an Agents folder in your workspace. You can also let the agent use the workspace's pinned skills by switching that on in Settings.
Connections
Agents can reach outside tools and data sources through connections. Add one on the agent's panel:
Some connections use a simple API key or token.
Others connect by signing in — add the connection, then click Connect to authorize it with the provider.
Each connection shows whether it's Connected, and you can reconnect at any time.
Workspace access
An agent always works in its home workspace. You can also grant it Read or Write access to other workspaces you administer — handy when it needs to pull from one place and write to another. It can only reach workspaces you give it.