Give your agents a clear role.
Define the work, the context and the review boundary before asking an agent to help.
Runs in your browser. No account needed.
A reusable role
Customer follow-up assistant- Instructions
- Use the conversation notes. Find one helpful next step.
- Review boundary
- Ask the owner before sending anything.
Give repeatable work a defined role
An agent is a saved role with instructions for a kind of work. The role explains what it should focus on, what information it needs and when it should ask a person to review the next step.
A customer follow-up assistant needs conversation notes and a clear boundary before sending anything. A support triage assistant needs the reported issue and a way to flag missing facts. Writing those expectations down makes the proposed work easier to judge.
Put it to work
- Customer follow-up assistant
- Read the supplied account context, identify an open question and prepare a draft for review.
- Support triage assistant
- Explain a suggested priority using the issue notes, and ask for information needed to investigate.
- Editorial planning assistant
- Check a draft against its audience brief and identify facts that still need review.
Explore a role before relying on its work
Explore agents, then follow the example there:
- Choose an example and review the role, instructions and fictional context in the creation flow.
- Save the agent, then edit its title, context and steps to make the expectations more specific.
- Select Review simulated run, then Run simulation to inspect the scripted scenario draft alongside your saved definition.
- Return after a reload to keep working with the same local example and creation chat.
How it fits in your workspace
Common questions
- What is the difference between a TulipFarm agent and an agent task?
- An agent is a saved role with instructions for a kind of work. An agent task is one specific request to an agent, with its own inputs and result. A hiring coordination agent describes how to prepare interviews; preparing Sam Taylor's interview is one task. In the Explore tour, sample agents and tasks are separate and aren't linked.
- Does Run simulation in the Explore tour call an AI model?
- No. Run simulation shows a fixed, fictional result next to the agent definition and steps you saved. It doesn't run a model, use tools or send messages.
Saving a demo agent creates a local definition, not an active agent in the hosted product. Simulations never send messages.
Explore a role before relying on its work
Start from an empty workspace and build something you can inspect and edit. Your changes stay in this browser.