Agents are reusable AI configurations for your workspace. They define a job once, then AI nodes can invoke that job consistently across workflows. Each run records the agent version and model used.
#Choose a starting point
Open Agents from the workspace sidebar.
- Platform agents are read-only starting points from Rills.
- Workspace agents are owned and managed by your team.
Duplicate a platform agent when its job is close to what you need. The copy is independent, so changes never affect the original or another workspace. Create a workspace agent for a distinct recurring job, such as categorizing replies, researching accounts, or drafting follow-ups.
#Use an agent in a workflow
In an AI node, select an agent, provide its inputs, and describe the task for that run. The node shows the agent’s output contract, tools, model summary, and estimated AI-credit cost. Its configuration remains read-only there so one workflow cannot silently change a shared job.
#Next
- Configure an agent: instructions, inputs, output, limits, context, and permissions
- Improve an agent: versions, feedback, evaluations, memory, and rollback