<aside> 💡

This page contains best practices to use the Context input when triggering agent runs.

</aside>

Introduction

The Utopian Labs agents are built to be as flexible as possible, giving you full control over input and output to maximize performance. The best way to influence the output of agent runs is to use the Context input.

The Context input can be seen as the system prompt, where you can add all kinds of additional information. It can be specifying previous interactions, specific qualifying factors that are relevant, or instructions on what an email should look like.

If you leave the Context input empty or provide minimal information, the agent will call upon it’s internal reasoning capabilities to create the Research Objective and figure out which tools to use to satisfy this objective. You can also specify which information you’re looking for to influence the Research Objective.

Generic best practices

• Determine how much creativity you’re looking for. The more details in the Context input, the less creative the agent will be.
• Use easy-to-understand language. The more fluffy or corporate terms, the harder it will be for the agent to understand. The easier the language, the granular the result.
• Be direct about what you want. The more direct, the less distraction for the agents.
• Write your context as you would write a briefing for a new employee. The agent has basic understanding about you and your company, but it doesn’t have the insight info you received from experience
• Be structured in your input, using bullets and short sentences.
  • Bonus: use XML tags to structure your input. For example; <copywriting_instructions> ... </copywriting_instructions>.

Examples

R1

If you want to do generic research with maximum creativity, you can leave the Context input empty. The Research Agent will then reason what the most interesting insights would be and starts actively researching these.

If you have a specific request that you’re looking for, you can mention that like this: