The System Prompt

The system prompt is the standing instruction the model reads before your first message, and keeps in view for the whole session. Where Thinking and Creativity change a single reply, the system prompt sets who the agent is and how it should behave throughout: its tone, what it should assume about you, the conventions it should follow. It is the most durable way to shape the agent.

It sits at the top of the conversation, collapsed, so you can check what the agent was told without it crowding the chat. Because it applies to every turn, a small change here ripples through everything the agent says next.

Presets

You do not have to write a system prompt from scratch. Loca ships a few starting points under Prompts in Settings: a plain assistant, a tool-focused agent, or none at all. Pick the one closest to how you want to work and edit the text to taste, rather than facing a blank box.

A good system prompt is specific about the things you would otherwise repeat in every message: the format you want answers in, the standards to hold, what to do when it is unsure. Anything you find yourself saying again and again is a candidate to move here once.

Helper prompts

A handful of smaller prompts drive the work you see around the edges of a session: naming a session from its first message, summarizing a memory, cleaning up a dictation, and the complexity check that routes a message between two models. They live in the same Prompts group as the system prompt, so the language Loca uses on your behalf is shaped in one place rather than scattered through the app.

How It Works

When the Core assembles a turn, the system prompt goes in first, ahead of the conversation history and anything the tools or memories add, so the model reads your standing instruction before it reads a single message. That position is why it carries across every turn at once: it is re-sent at the front of each request, not stored inside the model. The helper prompts work the same way, only for the Core’s own background jobs. Each one is a small, self-contained instruction the Core runs on its own when it needs a title, a summary, or a cleanup, which is what makes them editable in the same place as your main prompt: they are prompts, not hard-coded behavior.

Next, a prompt that runs on its own when you open the app: Greetings.