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

# System Prompt

> Define your agent's behavior — inline or by reference to a versioned prompt template

The system prompt is what tells the model *what kind of agent it is*. HasteKit lets you write it inline on the agent, or reference an immutable version of a [prompt template](/docs/gateway/agent-builder/prompts) from your project's prompt library so the same prompt can power many agents.

## Two ways to configure

### Use raw prompt text

Write the prompt directly on the agent. The editor supports multi-line text and syntax highlighting.

<Frame>
  <img src="https://mintcdn.com/hastekit/890s5wy2NHvxQLSN/images/agent-builder/system-prompt-raw.png?fit=max&auto=format&n=890s5wy2NHvxQLSN&q=85&s=ec2ecca0ceac7d3022ca25502a556aa9" alt="System Prompt Raw Text" width="1586" height="984" data-path="images/agent-builder/system-prompt-raw.png" />
</Frame>

Best when:

* The prompt is unique to this agent.
* You're iterating quickly and don't yet want to version it.

### Use prompt reference

Reference a prompt from your project's [Prompts](/docs/gateway/agent-builder/prompts) library.

1. Select **Reference**.
2. Pick a prompt from the **Select Prompt** dropdown.
3. Optionally pin to a specific version (or leave on `$LATEST`).

<Frame>
  <img src="https://mintcdn.com/hastekit/890s5wy2NHvxQLSN/images/agent-builder/system-prompt-reference.png?fit=max&auto=format&n=890s5wy2NHvxQLSN&q=85&s=ac5fc7a4f2b697221cb4fc9bc4f23be1" alt="System Prompt Reference" width="1586" height="772" data-path="images/agent-builder/system-prompt-reference.png" />
</Frame>

Each prompt version is **immutable** — referencing version `3` of a prompt guarantees the exact same text forever, even if a teammate later publishes version `4`. This is the recommended path for production agents.

Best when:

* Multiple agents share a prompt.
* You want to version the prompt independently of the agent.
* You want a non-engineer to be able to edit the prompt without touching the agent config.

## Variables

Prompt templates can use `{{variable}}` interpolation. Variables are resolved from the agent's runtime context (passed via the [chat UI's Add Context dialog](/docs/gateway/agent-builder/conversing-with-the-agent) or the HTTP request body). This lets one prompt template adapt to per-user or per-tenant data without spinning up a separate agent for each.