google/dotprompt

Dotprompt – Executable Prompt Templates for Generative AI

What it is – Dotprompt defines a file format (.prompt) that bundles a generative‑AI prompt together with all the information needed to run it: model selection, temperature, input schema, expected output format, and Handlebars‑style placeholders. The format is language‑agnostic, and Google ships core libraries for JavaScript/TypeScript, Python, Go, Rust, and Java, plus editor extensions (VS Code, JetBrains, Vim/Emacs) and web‑based editors (Monaco, CodeMirror).

Why it matters – Prompt engineering often lives in scattered code or ad‑hoc text files. By making a prompt a self‑contained, executable artifact you can:

• Validate inputs against a schema before sending them to the model.
• Declare the model and its parameters (e.g., gemini-2.5-pro, temperature) right in the file.
• Define the expected output format (JSON, XML, etc.) and schema, enabling automatic parsing and type‑checking.
• Reuse the same prompt across projects and languages without rewriting boilerplate.

Key components

Component	What it provides
Core libraries (dotprompt, dotpromptz, dotprompt-go, dotprompt-rs, dotprompt-java)	Parsers, validators, and runtime helpers that read a .prompt file, fill in Handlebars variables, call the configured model, and return typed results.
IDE extensions (VS Code, JetBrains, Vim, Emacs)	Syntax highlighting, schema‑aware autocompletion, and one‑click “run prompt” commands inside the editor.
Web editors (Monaco, CodeMirror)	Embeddable editors for SaaS tools or internal dashboards.
Tree‑sitter grammar	Precise parsing for language‑server features and custom tooling.
CLI (promptly)	Run a .prompt file from the terminal, useful for quick experiments or CI pipelines.

Typical workflow

1. Create a .prompt file – start with a YAML front‑matter block that declares model, input.schema, output.format, and output.schema. Write the prompt body using Handlebars ({{variable}}).
2. Add the library – e.g., npm i dotprompt (JS) or pip install dotpromptz (Python).
3. Load the prompt in code (ai.prompt("myPrompt") in Genkit, or the equivalent in other languages).
4. Call it with a plain object matching the input schema; the library handles validation, model invocation, and output parsing.
5. Iterate – because the prompt is a version‑controlled file, you can track changes, roll back, and share it across teams.

Example (excerpt from the README) – a prompt that extracts structured data from free‑form text:

---
model: googleai/gemini-2.5-pro
input:
  schema:
    text: string
output:
  format: json
  schema:
    name?: string, the full name of the person
    age?: number, the age of the person
    occupation?: string, the person's occupation
---

Extract the requested information from the given text. If a piece of information
is not present, omit that field from the output.

Text: {{text}}

Running this with the Genkit integration yields a typed JSON object like {"name":"John Doe","age":35,"occupation":"software engineer"}.

Getting started (JS/TS example shown in the README)

npm i genkit @genkit-ai/googleai   # install Genkit + Google AI provider

import { genkit } from "genkit";
import { googleAI } from "@genkit-ai/googleai";

const ai = genkit({ plugins: [googleAI()] });
const extractInfo = ai.prompt("extractInfo");

const {output} = await extractInfo({text: "John Doe is a 35‑year‑old software engineer..."});
console.log(output);

Set GOOGLE_API_KEY in the environment to authenticate.

Who should use it

• Developers building AI‑augmented applications – keep prompts tidy, versioned, and type‑safe.
• Prompt engineers / data scientists – experiment with model parameters and output schemas without touching application code.
• Teams with multi‑language stacks – the same .prompt file works in JS, Python, Go, Rust, and Java.
• Tool builders – embed the web editors or CLI into internal platforms.

Community & support

• Discord channel for real‑time help: https://discord.gg/qXt5zzQKpc
• Open‑source contributions welcomed; see GEMINI.md for commit conventions and testing guidelines.
• CI badges show that each language implementation is continuously tested.

---

Dotprompt turns prompts into first‑class, executable assets, making generative‑AI development more modular, reliable, and collaborative.