alchemy 0.2.5 → 0.3.0

package/src/ai/html-file.ts

@@ -0,0 +1,263 @@
+import { generateText } from "ai";
+import fs from "node:fs/promises";
+import path from "node:path";
+import type { Context } from "../context";
+import { Resource } from "../resource";
+import type { Secret } from "../secret";
+import { ignore } from "../util/ignore";
+import { type ModelConfig, createModel } from "./client";
+
+/**
+ * Properties for creating or updating an HTMLFile
+ */
+export interface HTMLFileProps {
+  /**
+   * Path to the HTML file
+   */
+  path: string;
+
+  /**
+   * Base URL for the OpenAI API
+   * @default 'https://api.openai.com/v1'
+   */
+  baseURL?: string;
+
+  /**
+   * Prompt for generating content
+   * Use alchemy template literals to include file context:
+   * @example
+   * prompt: await alchemy`
+   *   Generate an HTML page using:
+   *   ${alchemy.file("src/templates/base.html")}
+   * `
+   */
+  prompt: string;
+
+  /**
+   * System prompt for the model
+   * This is used to provide instructions to the model about how to format the response
+   * The default system prompt instructs the model to return HTML code inside ```html fences
+   * @default "You are an HTML code generator. Create HTML code based on the user's requirements. Your response MUST include only HTML code inside ```html fences. Do not include any other text, explanations, or multiple code blocks."
+   */
+  system?: string;
+
+  /**
+   * OpenAI API key to use for generating content
+   * If not provided, will use OPENAI_API_KEY environment variable
+   */
+  apiKey?: Secret;
+
+  /**
+   * Model configuration
+   */
+  model?: ModelConfig;
+
+  /**
+   * Temperature for controlling randomness in generation.
+   * Higher values (e.g., 0.8) make output more random,
+   * lower values (e.g., 0.2) make it more deterministic.
+   * @default 0.7
+   */
+  temperature?: number;
+}
+
+/**
+ * An HTML file that can be created, updated, and deleted
+ */
+export interface HTMLFile extends HTMLFileProps, Resource<"ai::HTMLFile"> {
+  /**
+   * Content of the HTML file
+   */
+  content: string;
+
+  /**
+   * Time at which the file was created
+   */
+  createdAt: number;
+
+  /**
+   * Time at which the file was last updated
+   */
+  updatedAt: number;
+}
+
+/**
+ * Default system prompt for HTML file generation
+ */
+const DEFAULT_HTML_SYSTEM_PROMPT =
+  "You are an HTML code generator. Create HTML code based on the user's requirements. Your response MUST include only HTML code inside ```html fences. Do not include any other text, explanations, or multiple code blocks.";
+
+/**
+ * Resource for generating HTML files using AI models.
+ * Extracts HTML code from between ```html fences and validates the response.
+ *
+ * @example
+ * // Create a simple landing page
+ * const landingPage = await HTMLFile("landing-page", {
+ *   path: "./public/index.html",
+ *   prompt: await alchemy`
+ *     Generate a modern landing page for a SaaS product with:
+ *     - Hero section with headline and call-to-action
+ *     - Features section with 3 key features
+ *     - Pricing section with 3 tiers
+ *     - Testimonials section with 2 customer quotes
+ *     - Contact form and footer
+ *   `,
+ *   model: {
+ *     id: "gpt-4o",
+ *     provider: "openai"
+ *   }
+ * });
+ *
+ * @example
+ * // Generate an HTML email template
+ * const emailTemplate = await HTMLFile("welcome-email", {
+ *   path: "./emails/welcome.html",
+ *   prompt: await alchemy`
+ *     Create an HTML email template for welcoming new users to our platform.
+ *     The email should include:
+ *     - Company logo and branding
+ *     - Personalized welcome message (use {{name}} placeholder)
+ *     - Three steps to get started
+ *     - Support contact information
+ *     - Unsubscribe footer
+ *
+ *     Make sure it's responsive and works in all major email clients.
+ *   `,
+ *   temperature: 0.2
+ * });
+ *
+ * @example
+ * // Generate an HTML component with custom system prompt
+ * const navComponent = await HTMLFile("navigation", {
+ *   path: "./components/nav.html",
+ *   prompt: await alchemy`
+ *     Create a responsive navigation component with:
+ *     - Logo in the left corner
+ *     - Navigation links: Home, Products, Services, About, Contact
+ *     - Mobile hamburger menu that expands/collapses
+ *     - Login/signup buttons on the right side
+ *     - Dark/light mode toggle
+ *   `,
+ *   system: "You are an expert HTML/CSS developer specializing in responsive components. Create a single HTML file inside ```html fences with no additional text. Use modern HTML5 semantic elements and inline CSS if needed.",
+ *   model: {
+ *     id: "claude-3-opus-20240229",
+ *     provider: "anthropic"
+ *   }
+ * });
+ */
+export const HTMLFile = Resource(
+  "ai::HTMLFile",
+  async function (
+    this: Context<HTMLFile>,
+    id: string,
+    props: HTMLFileProps,
+  ): Promise<HTMLFile> {
+    // Ensure directory exists
+    await fs.mkdir(path.dirname(props.path), { recursive: true });
+
+    if (this.phase === "delete") {
+      try {
+        await fs.unlink(props.path);
+      } catch (error: any) {
+        // Ignore if file doesn't exist
+        if (error.code !== "ENOENT") {
+          throw error;
+        }
+      }
+      return this.destroy();
+    }
+
+    // Use provided system prompt or default
+    const system = props.system || DEFAULT_HTML_SYSTEM_PROMPT;
+
+    // Generate initial content
+    const { text } = await generateText({
+      model: createModel(props),
+      prompt: props.prompt,
+      system,
+      providerOptions: props.model?.options,
+      ...(props.temperature === undefined
+        ? {}
+        : { temperature: props.temperature }),
+    });
+
+    // Extract and validate HTML code
+    let { code, error } = await extractHTMLCode(text);
+
+    // Re-prompt if there are validation errors
+    if (error) {
+      const errorSystem = `${system}\n\nERROR: ${error}\n\nPlease try again and ensure your response contains exactly one HTML code block inside \`\`\`html fences.`;
+
+      const { text: retryText } = await generateText({
+        model: createModel(props),
+        prompt: props.prompt,
+        system: errorSystem,
+        providerOptions: props.model?.options,
+        ...(props.temperature === undefined
+          ? {}
+          : { temperature: props.temperature }),
+      });
+
+      const retryResult = await extractHTMLCode(retryText);
+
+      if (retryResult.error) {
+        throw new Error(
+          `Failed to generate valid HTML code: ${retryResult.error}`,
+        );
+      }
+
+      code = retryResult.code;
+    }
+
+    if (this.phase === "update" && props.path !== this.props.path) {
+      await ignore("ENOENT", () => fs.unlink(this.props.path));
+    }
+
+    // Write content to file
+    await fs.writeFile(props.path, code);
+
+    // Get file stats for timestamps
+    const stats = await fs.stat(props.path);
+
+    // Return the resource
+    return this({
+      ...props,
+      content: code,
+      createdAt: stats.birthtimeMs,
+      updatedAt: stats.mtimeMs,
+    });
+  },
+);
+
+/**
+ * Extracts HTML code from between ```html fences
+ * Validates that exactly one HTML code block exists
+ *
+ * @param text The text to extract HTML code from
+ * @returns The extracted HTML code or error message
+ */
+async function extractHTMLCode(
+  text: string,
+): Promise<{ code: string; error?: string }> {
+  const htmlCodeRegex = /```html\s*([\s\S]*?)```/g;
+  const matches = Array.from(text.matchAll(htmlCodeRegex));
+
+  if (matches.length === 0) {
+    return {
+      code: "",
+      error:
+        "No HTML code block found in the response. Please include your code within ```html fences.",
+    };
+  }
+
+  if (matches.length > 1) {
+    return {
+      code: "",
+      error:
+        "Multiple HTML code blocks found in the response. Please provide exactly one code block within ```html fences.",
+    };
+  }
+
+  return { code: matches[0][1].trim() };
+}

package/src/ai/index.ts

@@ -1,3 +1,10 @@
 export * from "./ark";
+export * from "./astro-file";
+export * from "./css-file";
 export * from "./data";
 export * from "./document";
+export * from "./html-file";
+export * from "./json-file";
+export * from "./typescript-file";
+export * from "./vue-file";
+export * from "./yaml-file";

package/src/ai/json-file.ts

@@ -0,0 +1,349 @@
+import { generateObject, generateText } from "ai";
+import type { JsonSchema, Type, type } from "arktype";
+import fs from "node:fs/promises";
+import path from "node:path";
+import type { Context } from "../context";
+import { Resource } from "../resource";
+import type { Secret } from "../secret";
+import { ignore } from "../util/ignore";
+import { ark } from "./ark";
+import { type ModelConfig, createModel } from "./client";
+
+/**
+ * Properties for creating or updating a JSONFile
+ */
+export interface JSONFileProps<
+  T extends Type<any, any> | undefined = undefined,
+> {
+  /**
+   * Path to the JSON file
+   */
+  path: string;
+
+  /**
+   * Optional ArkType schema to validate and structure the generated JSON
+   * When provided, the resource will use generateObject with schema validation
+   * When not provided, it will extract JSON from between ```json fences
+   */
+  schema?: T;
+
+  /**
+   * Base URL for the OpenAI API
+   * @default 'https://api.openai.com/v1'
+   */
+  baseURL?: string;
+
+  /**
+   * Prompt for generating content
+   * Use alchemy template literals to include file context:
+   * @example
+   * prompt: await alchemy`
+   *   Generate a JSON configuration for:
+   *   ${alchemy.file("src/config.ts")}
+   * `
+   */
+  prompt: string;
+
+  /**
+   * System prompt for the model
+   * This is used to provide instructions to the model about how to format the response
+   * @default Depends on whether schema is provided
+   */
+  system?: string;
+
+  /**
+   * OpenAI API key to use for generating content
+   * If not provided, will use OPENAI_API_KEY environment variable
+   */
+  apiKey?: Secret;
+
+  /**
+   * Model configuration
+   */
+  model?: ModelConfig;
+
+  /**
+   * Temperature for controlling randomness in generation.
+   * Higher values (e.g., 0.8) make output more random,
+   * lower values (e.g., 0.2) make it more deterministic.
+   * @default 0.7
+   */
+  temperature?: number;
+
+  /**
+   * Whether to pretty-print the JSON with indentation
+   * @default true
+   */
+  pretty?: boolean;
+
+  /**
+   * Number of spaces to use for indentation when pretty-printing
+   * @default 2
+   */
+  indent?: number;
+}
+
+/**
+ * A JSON file that can be created, updated, and deleted
+ */
+export interface JSONFile<T = any>
+  extends Omit<JSONFileProps, "schema">,
+    Resource<"ai::JSONFile"> {
+  /**
+   * Content of the JSON file as a string
+   */
+  content: string;
+
+  /**
+   * Parsed JSON object
+   */
+  json: T;
+
+  /**
+   * Schema used to validate the JSON (if provided)
+   */
+  schema?: JsonSchema;
+
+  /**
+   * Time at which the file was created
+   */
+  createdAt: number;
+
+  /**
+   * Time at which the file was last updated
+   */
+  updatedAt: number;
+}
+
+/**
+ * Default system prompt for JSON file generation without schema
+ */
+const DEFAULT_JSON_SYSTEM_PROMPT =
+  "You are a JSON generator. Create valid JSON based on the user's requirements. Your response MUST include only JSON inside ```json fences. Do not include any other text, explanations, or multiple code blocks.";
+
+/**
+ * Resource for generating JSON files using AI models.
+ * Can operate in two modes:
+ * 1. With schema: Uses generateObject with type validation
+ * 2. Without schema: Extracts JSON from between ```json fences
+ *
+ * @example
+ * // Generate a configuration file with freeform JSON
+ * const config = await JSONFile("app-config", {
+ *   path: "./config/app.json",
+ *   prompt: await alchemy`
+ *     Generate a configuration for a web application with:
+ *     - Server settings (port, host, timeout)
+ *     - Database connection details (redact any passwords)
+ *     - Logging configuration
+ *     - Feature flags
+ *   `,
+ *   model: {
+ *     id: "gpt-4o",
+ *     provider: "openai"
+ *   }
+ * });
+ *
+ * @example
+ * // Generate JSON with schema validation
+ * import { type } from "arktype";
+ *
+ * const userSchema = type({
+ *   users: [{
+ *     id: "string",
+ *     name: "string",
+ *     email: "string",
+ *     role: "'admin' | 'user' | 'guest'",
+ *     permissions: "string[]",
+ *     active: "boolean"
+ *   }]
+ * });
+ *
+ * const userData = await JSONFile("user-data", {
+ *   path: "./data/users.json",
+ *   schema: userSchema,
+ *   prompt: "Generate sample user data for an application with various roles and permissions",
+ *   temperature: 0.2
+ * });
+ *
+ * // Type-safe access to the generated data
+ * console.log(userData.json.users[0].role); // Typed as 'admin' | 'user' | 'guest'
+ *
+ * @example
+ * // Generate API mock data with custom system prompt
+ * const apiMock = await JSONFile("api-mock", {
+ *   path: "./mocks/products-api.json",
+ *   prompt: await alchemy`
+ *     Create mock data for a product catalog API response with:
+ *     - 10 products with different categories
+ *     - Each product should have id, name, price, category, inventory, and image_url
+ *     - Include pagination metadata (total, page, limit)
+ *   `,
+ *   system: "You are an API design expert. Create realistic mock JSON data that follows REST API best practices. Your response must be valid JSON inside ```json fences.",
+ *   model: {
+ *     id: "claude-3-opus-20240229",
+ *     provider: "anthropic"
+ *   },
+ *   pretty: true,
+ *   indent: 4
+ * });
+ */
+export const JSONFile = Resource("ai::JSONFile", async function <
+  const T extends Type<any, any> | undefined = undefined,
+>(this: Context<JSONFile<T extends Type<any, any> ? type.infer<T> : any>>, id: string, props: JSONFileProps<T>): Promise<
+  JSONFile<T extends Type<any, any> ? type.infer<T> : any>
+> {
+  // Ensure directory exists
+  await fs.mkdir(path.dirname(props.path), { recursive: true });
+
+  if (this.phase === "delete") {
+    try {
+      await fs.unlink(props.path);
+    } catch (error: any) {
+      // Ignore if file doesn't exist
+      if (error.code !== "ENOENT") {
+        throw error;
+      }
+    }
+    return this.destroy();
+  }
+
+  // Determine if we should use pretty printing
+  const pretty = props.pretty !== false;
+  const indent = props.indent ?? 2;
+
+  let jsonContent: string;
+  let jsonObject: any;
+
+  // Check if schema is provided
+  if (props.schema) {
+    // Use schema-based generation
+    const { object } = await generateObject({
+      model: createModel(props),
+      schema: ark.schema<type.infer<typeof props.schema>>(props.schema),
+      providerOptions: props.model?.options,
+      system:
+        props.system ||
+        "Generate a valid JSON object based on the provided requirements.",
+      prompt: props.prompt,
+      ...(props.temperature === undefined
+        ? {}
+        : { temperature: props.temperature }),
+    });
+
+    jsonObject = object;
+    jsonContent = pretty
+      ? JSON.stringify(jsonObject, null, indent)
+      : JSON.stringify(jsonObject);
+  } else {
+    // Use fence-based extraction
+    // Use provided system prompt or default
+    const system = props.system || DEFAULT_JSON_SYSTEM_PROMPT;
+
+    // Generate initial content
+    const { text } = await generateText({
+      model: createModel(props),
+      prompt: props.prompt,
+      system,
+      providerOptions: props.model?.options,
+      ...(props.temperature === undefined
+        ? {}
+        : { temperature: props.temperature }),
+    });
+
+    // Extract and validate JSON content
+    let { content, error } = await extractJSONContent(text);
+
+    // Re-prompt if there are validation errors
+    if (error) {
+      const errorSystem = `${system}\n\nERROR: ${error}\n\nPlease try again and ensure your response contains exactly one valid JSON block inside \`\`\`json fences.`;
+
+      const { text: retryText } = await generateText({
+        model: createModel(props),
+        prompt: props.prompt,
+        system: errorSystem,
+        providerOptions: props.model?.options,
+        ...(props.temperature === undefined
+          ? {}
+          : { temperature: props.temperature }),
+      });
+
+      const retryResult = await extractJSONContent(retryText);
+
+      if (retryResult.error) {
+        throw new Error(`Failed to generate valid JSON: ${retryResult.error}`);
+      }
+
+      content = retryResult.content;
+    }
+
+    // Parse JSON
+    jsonObject = JSON.parse(content);
+
+    // Format with or without indentation based on pretty option
+    jsonContent = pretty ? JSON.stringify(jsonObject, null, indent) : content;
+  }
+
+  if (this.phase === "update" && props.path !== this.props.path) {
+    await ignore("ENOENT", () => fs.unlink(this.props.path));
+  }
+
+  // Write content to file
+  await fs.writeFile(props.path, jsonContent);
+
+  // Get file stats for timestamps
+  const stats = await fs.stat(props.path);
+
+  // Return the resource
+  return this({
+    ...props,
+    schema: props.schema,
+    content: jsonContent,
+    json: jsonObject,
+    createdAt: stats.birthtimeMs,
+    updatedAt: stats.mtimeMs,
+  });
+});
+
+/**
+ * Extracts JSON content from between ```json fences
+ * Validates that exactly one JSON code block exists
+ *
+ * @param text The text to extract JSON from
+ * @returns The extracted JSON or error message
+ */
+async function extractJSONContent(
+  text: string,
+): Promise<{ content: string; error?: string }> {
+  const jsonCodeRegex = /```json\s*([\s\S]*?)```/g;
+  const matches = Array.from(text.matchAll(jsonCodeRegex));
+
+  if (matches.length === 0) {
+    return {
+      content: "",
+      error:
+        "No JSON code block found in the response. Please include your JSON within ```json fences.",
+    };
+  }
+
+  if (matches.length > 1) {
+    return {
+      content: "",
+      error:
+        "Multiple JSON code blocks found in the response. Please provide exactly one JSON block within ```json fences.",
+    };
+  }
+
+  const content = matches[0][1].trim();
+
+  // Validate JSON can be parsed
+  try {
+    JSON.parse(content);
+    return { content };
+  } catch (e) {
+    return {
+      content: "",
+      error: `Invalid JSON: ${(e as Error).message}. Please provide valid JSON syntax.`,
+    };
+  }
+}