> ## 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. # Prompts > Versioned, reusable prompt templates with `{{variable}}` interpolation The **Prompts** library is your project's source of truth for system prompts. Author a prompt once, version it like code, and reference specific versions from any number of agents — so changing a prompt doesn't require touching every agent that uses it. Prompts library ## Why prompts deserve their own library A prompt that lives inline on an agent is fine while you're iterating, but it has costs: * **Duplication** — every agent that needs similar instructions has its own copy that drifts. * **No version control** — there's no record of what the prompt said when an old run happened. * **Mixed ownership** — engineers edit agent configs, but prompt-writers want to edit prompts. A prompt in the library has: * **Immutable versions** — every saved version is frozen. Reference `my-prompt@3` and you'll get the exact same text forever. * **Shared use** — one prompt powers many agents. ## Creating a prompt 1. Open **Project → Prompts**. 2. Click **Create Prompt**. 3. Give it a **Name** (unique per project; used to reference the prompt from agents) and an optional **Description**. 4. Write the prompt text. Use `{{variable}}` for any dynamic substitutions. 5. Click **Save**. This creates version 1. ## Versioning Each time you save edits to an existing prompt, it updates the `$LATEST` version. You can create a immutable version from `$LATEST` through the **Versions → Create Version**. The **Versions** tab lists every version with its creation timestamp; you can view any historical version verbatim. A prompt-reference on an agent ([System Prompt tab](/docs/gateway/agent-builder/system-prompt)) pins to a specific version. If you want an agent to always use the newest version, pin it to `$LATEST`; if you want stable behaviour, pin to an explicit number. ## Variables A template body can reference `{{name}}` anywhere. At runtime, variables resolve from the request's **context** object — the same object you set in the [Chat UI's Add Context dialog](/docs/gateway/agent-builder/conversing-with-the-agent) or pass in the HTTP request body. Example: ``` You are a support agent for {{company_name}}. The customer you're helping is {{customer_name}} ({{customer_tier}} tier). Always be polite and reference their tier when relevant. ``` If `{{customer_tier}}` isn't provided in context at runtime, it's left blank — variables are optional unless the prompt logic requires them. ## When to reference a prompt vs write it inline * **Inline** — quick iteration on a single agent, throwaway experiments, prompt is hyper-specific to this one agent. * **Reference** — production, shared logic, anything you want to version independently of the agent definition. ## Related * [System Prompt](/docs/gateway/agent-builder/system-prompt) — the agent-side tab where you pick "raw text" or "prompt reference." * [Versioning](/docs/gateway/agent-builder/versioning) — the same immutable-versions model applies to agents themselves.