> ## Documentation Index
> Fetch the complete documentation index at: https://docs.retellai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Single prompt vs. conversation flow

> Compare single prompt and conversation flow agents: when each fits, the control vs. design tradeoff, and why to start with a single prompt agent first.

A single prompt gives the model instructions for the whole task and lets it decide how to conduct the conversation and use its tools. Conversation flow divides the task into nodes, each with its own instructions or actions. You connect the nodes and define the conditions for moving between them.

For your first version, we recommend starting with a single prompt agent. Switch to conversation flow only when a single prompt agent can no longer meet your requirements.

## When to use a single prompt agent

Use a single prompt when the task can be described clearly in one prompt and the model can decide how to carry it out. Having multiple steps does not, by itself, require conversation flow.

If your single prompt agent already works reliably, keep it as a single prompt agent. Switching to conversation flow requires restructuring the agent into nodes and transitions, adding design and maintenance work without necessarily improving its behavior.

## When to use a conversation flow agent

Conversation flow is useful when you need one or more of the following:

* **More deterministic execution.** Defining the order of nodes and the actions they execute can give you tighter control over execution order and outcomes. This can matter for compliance requirements. Some decisions still depend on the model, so using conversation flow does not make every part of the agent deterministic.
* **Less context per node.** As a single prompt grows, the model may struggle to follow all of its instructions. Increased hallucinations and instruction-following problems can be signs of this. First simplify and shorten the prompt. If the necessary instructions and tool definitions cannot be reduced enough, consider rebuilding the agent as a conversation flow. You can then divide the task among nodes, each with the instructions and tools needed for that part of the task.

## The tradeoff

Single prompt agents require less design work and leave more decisions to the model. Conversation flow provides more explicit control, but introduces node boundaries and transition conditions that need careful design.

Conversation flow is not automatically more reliable. Unnecessary nodes and transitions can make a simple conversation harder to handle. Use the additional structure where its benefits justify it, and keep the conversation flow as simple as the task allows.

Once you choose an agent type, follow the [single prompt best practices](/build/single-multi-prompt/write-single-prompt) or [conversation flow best practices](/build/conversation-flow/best-practices). For multi-prompt agents, see the [legacy multi-prompt guide](/build/single-multi-prompt/write-multi-prompt).