mulmocast-preprocessor 0.1.2 → 0.2.1

package/lib/core/ai/command/query/prompts.js

@@ -0,0 +1,88 @@
+import { getLanguageName, buildScriptContent } from "../../llm.js";
+/**
+ * Default system prompt for query
+ */
+export const DEFAULT_SYSTEM_PROMPT = `You are answering questions based on the content provided.
+- Answer based ONLY on the information in the provided content
+- If the answer cannot be found in the content, say so clearly
+- Be concise and direct in your answers
+- Do not make up information that is not in the content`;
+/**
+ * Get system prompt based on options
+ */
+export const getSystemPrompt = (options) => {
+    if (options.systemPrompt) {
+        return options.systemPrompt;
+    }
+    const basePrompt = DEFAULT_SYSTEM_PROMPT;
+    // Add language instruction if specified
+    if (options.lang) {
+        const langName = getLanguageName(options.lang);
+        return `${basePrompt}\n- IMPORTANT: Write the answer in ${langName}`;
+    }
+    return basePrompt;
+};
+/**
+ * Build user prompt from script and question
+ */
+export const buildUserPrompt = (script, question) => {
+    const parts = [];
+    // Add common script content (title, language, sections with beats)
+    parts.push(buildScriptContent(script));
+    parts.push("---");
+    parts.push("");
+    parts.push(`Question: ${question}`);
+    parts.push("");
+    parts.push("Answer:");
+    return parts.join("\n");
+};
+/**
+ * Default system prompt for interactive query
+ */
+export const DEFAULT_INTERACTIVE_SYSTEM_PROMPT = `You are answering questions based on the content provided.
+- Answer based ONLY on the information in the provided content
+- If the answer cannot be found in the content, say so clearly
+- Be concise and direct in your answers
+- Do not make up information that is not in the content
+- You may reference previous conversation when answering follow-up questions`;
+/**
+ * Get system prompt for interactive mode
+ */
+export const getInteractiveSystemPrompt = (options) => {
+    if (options.systemPrompt) {
+        return options.systemPrompt;
+    }
+    const basePrompt = DEFAULT_INTERACTIVE_SYSTEM_PROMPT;
+    if (options.lang) {
+        const langName = getLanguageName(options.lang);
+        return `${basePrompt}\n- IMPORTANT: Write the answer in ${langName}`;
+    }
+    return basePrompt;
+};
+/**
+ * Build user prompt with conversation history for interactive mode
+ */
+export const buildInteractiveUserPrompt = (script, question, history) => {
+    const parts = [];
+    // Add common script content (title, language, sections with beats)
+    parts.push(buildScriptContent(script));
+    parts.push("---");
+    parts.push("");
+    // Add conversation history if exists
+    if (history.length > 0) {
+        parts.push("Previous conversation:");
+        history.forEach((msg) => {
+            if (msg.role === "user") {
+                parts.push(`Q: ${msg.content}`);
+            }
+            else {
+                parts.push(`A: ${msg.content}`);
+            }
+        });
+        parts.push("");
+    }
+    parts.push(`Current question: ${question}`);
+    parts.push("");
+    parts.push("Answer:");
+    return parts.join("\n");
+};

package/lib/core/ai/command/summarize/index.d.ts

@@ -0,0 +1,8 @@
+import type { ExtendedScript } from "../../../../types/index.js";
+import type { SummarizeOptions, SummarizeResult } from "../../../../types/summarize.js";
+/**
+ * Main summarize function - generates a summary of the entire script
+ */
+export declare const summarizeScript: (script: ExtendedScript, options?: Partial<SummarizeOptions>) => Promise<SummarizeResult>;
+export type { SummarizeOptions, SummarizeResult, LLMProvider, SummarizeFormat } from "../../../../types/summarize.js";
+export { summarizeOptionsSchema, llmProviderSchema, summarizeFormatSchema } from "../../../../types/summarize.js";

package/lib/core/ai/command/summarize/index.js

@@ -0,0 +1,33 @@
+import { summarizeOptionsSchema } from "../../../../types/summarize.js";
+import { executeLLM, filterScript } from "../../llm.js";
+import { buildUserPrompt, getSystemPrompt } from "./prompts.js";
+/**
+ * Main summarize function - generates a summary of the entire script
+ */
+export const summarizeScript = async (script, options = {}) => {
+    // Validate and apply defaults
+    const validatedOptions = summarizeOptionsSchema.parse(options);
+    // Filter script if section/tags specified
+    const filteredScript = filterScript(script, validatedOptions);
+    const scriptTitle = script.title || "Untitled";
+    if (filteredScript.beats.length === 0) {
+        return {
+            summary: "No content to summarize.",
+            format: validatedOptions.format,
+            scriptTitle,
+            beatCount: 0,
+        };
+    }
+    // Build prompts
+    const systemPrompt = getSystemPrompt(validatedOptions);
+    const userPrompt = buildUserPrompt(filteredScript, validatedOptions);
+    // Execute LLM
+    const summary = await executeLLM(systemPrompt, userPrompt, validatedOptions, `Summarizing script "${script.title}" with ${validatedOptions.provider}... Beats: ${filteredScript.beats.length}, Format: ${validatedOptions.format}`);
+    return {
+        summary,
+        format: validatedOptions.format,
+        scriptTitle,
+        beatCount: filteredScript.beats.length,
+    };
+};
+export { summarizeOptionsSchema, llmProviderSchema, summarizeFormatSchema } from "../../../../types/summarize.js";

package/lib/core/ai/command/summarize/prompts.d.ts

@@ -0,0 +1,18 @@
+import type { SummarizeOptions } from "../../../../types/summarize.js";
+import type { ExtendedScript } from "../../../../types/index.js";
+/**
+ * Default system prompt for text summary
+ */
+export declare const DEFAULT_SYSTEM_PROMPT_TEXT = "You are creating a summary based on the content provided.\n- Extract and explain the actual information and knowledge from the content\n- Do NOT describe what the presentation/script is about (avoid phrases like \"this presentation explains...\" or \"the script describes...\")\n- Write as if you are directly explaining the topic to the reader\n- Be concise and informative\n- Output plain text only";
+/**
+ * Default system prompt for markdown summary
+ */
+export declare const DEFAULT_SYSTEM_PROMPT_MARKDOWN = "You are creating a summary based on the content provided.\n- Extract and explain the actual information and knowledge from the content\n- Do NOT describe what the presentation/script is about (avoid phrases like \"this presentation explains...\" or \"the script describes...\")\n- Write as if you are directly explaining the topic to the reader\n- Use markdown formatting (headers, bullet points, etc.)\n- Include a title, key points, and conclusion\n- Output well-formatted markdown";
+/**
+ * Build user prompt from entire script
+ */
+export declare const buildUserPrompt: (script: ExtendedScript, options: SummarizeOptions) => string;
+/**
+ * Get system prompt based on format and language
+ */
+export declare const getSystemPrompt: (options: SummarizeOptions) => string;

package/lib/core/ai/command/summarize/prompts.js

@@ -0,0 +1,50 @@
+import { getLanguageName, buildScriptContent } from "../../llm.js";
+/**
+ * Default system prompt for text summary
+ */
+export const DEFAULT_SYSTEM_PROMPT_TEXT = `You are creating a summary based on the content provided.
+- Extract and explain the actual information and knowledge from the content
+- Do NOT describe what the presentation/script is about (avoid phrases like "this presentation explains..." or "the script describes...")
+- Write as if you are directly explaining the topic to the reader
+- Be concise and informative
+- Output plain text only`;
+/**
+ * Default system prompt for markdown summary
+ */
+export const DEFAULT_SYSTEM_PROMPT_MARKDOWN = `You are creating a summary based on the content provided.
+- Extract and explain the actual information and knowledge from the content
+- Do NOT describe what the presentation/script is about (avoid phrases like "this presentation explains..." or "the script describes...")
+- Write as if you are directly explaining the topic to the reader
+- Use markdown formatting (headers, bullet points, etc.)
+- Include a title, key points, and conclusion
+- Output well-formatted markdown`;
+/**
+ * Build user prompt from entire script
+ */
+export const buildUserPrompt = (script, options) => {
+    const parts = [];
+    // Add common script content (title, language, sections with beats)
+    parts.push(buildScriptContent(script));
+    // Add target length if specified
+    if (options.targetLengthChars) {
+        parts.push(`Target summary length: approximately ${options.targetLengthChars} characters`);
+    }
+    parts.push("");
+    parts.push("Based on the above content, explain the topic directly to the reader:");
+    return parts.join("\n");
+};
+/**
+ * Get system prompt based on format and language
+ */
+export const getSystemPrompt = (options) => {
+    if (options.systemPrompt) {
+        return options.systemPrompt;
+    }
+    const basePrompt = options.format === "markdown" ? DEFAULT_SYSTEM_PROMPT_MARKDOWN : DEFAULT_SYSTEM_PROMPT_TEXT;
+    // Add language instruction if specified
+    if (options.lang) {
+        const langName = getLanguageName(options.lang);
+        return `${basePrompt}\n- IMPORTANT: Write the output in ${langName}`;
+    }
+    return basePrompt;
+};

package/lib/core/ai/llm.d.ts

@@ -0,0 +1,61 @@
+import type { ExtendedScript } from "../../types/index.js";
+import type { LLMProvider } from "../../types/summarize.js";
+/**
+ * Base options for LLM operations
+ */
+export interface BaseLLMOptions {
+    provider: LLMProvider;
+    model?: string;
+    temperature?: number;
+    maxTokens?: number;
+    lang?: string;
+    systemPrompt?: string;
+    verbose?: boolean;
+    section?: string;
+    tags?: string[];
+}
+/**
+ * Provider configuration
+ */
+export interface ProviderConfig {
+    agentName: string;
+    defaultModel: string;
+    keyName: string;
+    maxTokens?: number;
+}
+/**
+ * Get provider configuration
+ */
+export declare const getProviderConfig: (provider: LLMProvider) => ProviderConfig;
+/**
+ * Get API key for provider
+ */
+export declare const getProviderApiKey: (provider: LLMProvider) => string | undefined;
+/**
+ * Filter script based on options (section, tags)
+ */
+export declare const filterScript: (script: ExtendedScript, options: BaseLLMOptions) => ExtendedScript;
+/**
+ * Get language name from code
+ */
+export declare const getLanguageName: (langCode: string) => string;
+/**
+ * Build script content for user prompt (common part)
+ */
+export declare const buildScriptContent: (script: ExtendedScript) => string;
+/**
+ * Command execution result
+ */
+export interface CommandResult {
+    text: string;
+    scriptTitle: string;
+    beatCount: number;
+}
+/**
+ * Execute a command (summarize, query, etc.) with common logic
+ */
+export declare const executeCommand: <T extends BaseLLMOptions>(script: ExtendedScript, options: T, getSystemPrompt: (opts: T) => string, buildUserPrompt: (script: ExtendedScript) => string, verboseMessage: string) => Promise<CommandResult | null>;
+/**
+ * Execute LLM call with GraphAI
+ */
+export declare const executeLLM: (systemPrompt: string, userPrompt: string, options: BaseLLMOptions, verboseMessage?: string) => Promise<string>;

package/lib/core/ai/llm.js

@@ -0,0 +1,191 @@
+import dotenv from "dotenv";
+import { GraphAI, GraphAILogger } from "graphai";
+import * as vanillaAgents from "@graphai/vanilla";
+import { openAIAgent } from "@graphai/openai_agent";
+import { anthropicAgent } from "@graphai/anthropic_agent";
+import { groqAgent } from "@graphai/groq_agent";
+import { geminiAgent } from "@graphai/gemini_agent";
+import { filterBySection, filterByTags } from "../preprocessing/filter.js";
+dotenv.config({ quiet: true });
+const agents = vanillaAgents.default ?? vanillaAgents;
+const provider2Agent = {
+    openai: {
+        agentName: "openAIAgent",
+        defaultModel: "gpt-4o-mini",
+        keyName: "OPENAI_API_KEY",
+        maxTokens: 4096,
+    },
+    anthropic: {
+        agentName: "anthropicAgent",
+        defaultModel: "claude-sonnet-4-20250514",
+        keyName: "ANTHROPIC_API_KEY",
+        maxTokens: 4096,
+    },
+    groq: {
+        agentName: "groqAgent",
+        defaultModel: "llama-3.1-8b-instant",
+        keyName: "GROQ_API_KEY",
+        maxTokens: 4096,
+    },
+    gemini: {
+        agentName: "geminiAgent",
+        defaultModel: "gemini-2.0-flash",
+        keyName: "GEMINI_API_KEY",
+        maxTokens: 4096,
+    },
+};
+/**
+ * Get provider configuration
+ */
+export const getProviderConfig = (provider) => {
+    return provider2Agent[provider];
+};
+/**
+ * Get API key for provider
+ */
+export const getProviderApiKey = (provider) => {
+    const config = provider2Agent[provider];
+    return process.env[config.keyName];
+};
+/**
+ * Create GraphAI graph for LLM call
+ */
+const createLLMGraph = (agentName) => ({
+    version: 0.5,
+    nodes: {
+        systemPrompt: {},
+        userPrompt: {},
+        model: {},
+        temperature: {},
+        maxTokens: {},
+        llmCall: {
+            agent: agentName,
+            inputs: {
+                system: ":systemPrompt",
+                prompt: ":userPrompt",
+                model: ":model",
+                temperature: ":temperature",
+                max_tokens: ":maxTokens",
+            },
+        },
+        result: {
+            isResult: true,
+            agent: "copyAgent",
+            inputs: {
+                text: ":llmCall.text",
+            },
+        },
+    },
+});
+/**
+ * Filter script based on options (section, tags)
+ */
+export const filterScript = (script, options) => {
+    const afterSection = options.section ? filterBySection(script, options.section) : script;
+    const afterTags = options.tags && options.tags.length > 0 ? filterByTags(afterSection, options.tags) : afterSection;
+    return afterTags;
+};
+/**
+ * Get language name from code
+ */
+export const getLanguageName = (langCode) => {
+    const langMap = {
+        ja: "Japanese",
+        en: "English",
+        zh: "Chinese",
+        ko: "Korean",
+        fr: "French",
+        de: "German",
+        es: "Spanish",
+        it: "Italian",
+        pt: "Portuguese",
+        ru: "Russian",
+    };
+    return langMap[langCode] || langCode;
+};
+/**
+ * Build script content for user prompt (common part)
+ */
+export const buildScriptContent = (script) => {
+    const parts = [];
+    // Add script metadata
+    parts.push(`# Script: ${script.title}`);
+    parts.push(`Language: ${script.lang}`);
+    parts.push("");
+    // Collect all text from beats grouped by section
+    const sections = new Map();
+    script.beats.forEach((beat, index) => {
+        const text = beat.text || "";
+        if (!text.trim())
+            return;
+        const section = beat.meta?.section || "main";
+        if (!sections.has(section)) {
+            sections.set(section, []);
+        }
+        sections.get(section).push(`[${index}] ${text}`);
+    });
+    // Output by section
+    sections.forEach((texts, section) => {
+        parts.push(`## Section: ${section}`);
+        texts.forEach((t) => parts.push(t));
+        parts.push("");
+    });
+    return parts.join("\n");
+};
+/**
+ * Execute a command (summarize, query, etc.) with common logic
+ */
+export const executeCommand = async (script, options, getSystemPrompt, buildUserPrompt, verboseMessage) => {
+    const filteredScript = filterScript(script, options);
+    const scriptTitle = script.title || "Untitled";
+    if (filteredScript.beats.length === 0) {
+        return null;
+    }
+    const systemPrompt = getSystemPrompt(options);
+    const userPrompt = buildUserPrompt(filteredScript);
+    const text = await executeLLM(systemPrompt, userPrompt, options, verboseMessage);
+    return {
+        text,
+        scriptTitle,
+        beatCount: filteredScript.beats.length,
+    };
+};
+/**
+ * Execute LLM call with GraphAI
+ */
+export const executeLLM = async (systemPrompt, userPrompt, options, verboseMessage) => {
+    const providerConfig = getProviderConfig(options.provider);
+    const apiKey = getProviderApiKey(options.provider);
+    if (!apiKey) {
+        throw new Error(`API key not found for provider "${options.provider}". Please set the ${providerConfig.keyName} environment variable.`);
+    }
+    // Build GraphAI config
+    const config = {
+        openAIAgent: { apiKey: process.env.OPENAI_API_KEY },
+        anthropicAgent: { apiKey: process.env.ANTHROPIC_API_KEY },
+        groqAgent: { apiKey: process.env.GROQ_API_KEY },
+        geminiAgent: { apiKey: process.env.GEMINI_API_KEY },
+    };
+    // Create GraphAI instance
+    const graph = new GraphAI(createLLMGraph(providerConfig.agentName), {
+        ...agents,
+        openAIAgent,
+        anthropicAgent,
+        groqAgent,
+        geminiAgent,
+    }, { config });
+    if (options.verbose && verboseMessage) {
+        GraphAILogger.info(verboseMessage);
+    }
+    // Inject values
+    graph.injectValue("systemPrompt", systemPrompt);
+    graph.injectValue("userPrompt", userPrompt);
+    graph.injectValue("model", options.model ?? providerConfig.defaultModel);
+    graph.injectValue("temperature", options.temperature ?? 0.7);
+    graph.injectValue("maxTokens", options.maxTokens ?? providerConfig.maxTokens ?? 2048);
+    // Run graph
+    const graphResult = await graph.run();
+    // Extract text from result node
+    const resultNode = graphResult.result;
+    return resultNode?.text || "";
+};

package/lib/core/llm/index.d.ts

@@ -0,0 +1,45 @@
+import type { ExtendedScript } from "../../types/index.js";
+import type { LLMProvider } from "../../types/summarize.js";
+/**
+ * Base options for LLM operations
+ */
+export interface BaseLLMOptions {
+    provider: LLMProvider;
+    model?: string;
+    temperature?: number;
+    maxTokens?: number;
+    lang?: string;
+    systemPrompt?: string;
+    verbose?: boolean;
+    section?: string;
+    tags?: string[];
+}
+/**
+ * Provider configuration
+ */
+export interface ProviderConfig {
+    agentName: string;
+    defaultModel: string;
+    keyName: string;
+    maxTokens?: number;
+}
+/**
+ * Get provider configuration
+ */
+export declare const getProviderConfig: (provider: LLMProvider) => ProviderConfig;
+/**
+ * Get API key for provider
+ */
+export declare const getProviderApiKey: (provider: LLMProvider) => string | undefined;
+/**
+ * Filter script based on options (section, tags)
+ */
+export declare const filterScript: (script: ExtendedScript, options: BaseLLMOptions) => ExtendedScript;
+/**
+ * Get language name from code
+ */
+export declare const getLanguageName: (langCode: string) => string;
+/**
+ * Execute LLM call with GraphAI
+ */
+export declare const executeLLM: (systemPrompt: string, userPrompt: string, options: BaseLLMOptions, verboseMessage?: string) => Promise<string>;

package/lib/core/llm/index.js

@@ -0,0 +1,144 @@
+import dotenv from "dotenv";
+import { GraphAI, GraphAILogger } from "graphai";
+import * as vanillaAgents from "@graphai/vanilla";
+import { openAIAgent } from "@graphai/openai_agent";
+import { anthropicAgent } from "@graphai/anthropic_agent";
+import { groqAgent } from "@graphai/groq_agent";
+import { geminiAgent } from "@graphai/gemini_agent";
+import { filterBySection, filterByTags } from "../filter.js";
+dotenv.config({ quiet: true });
+const agents = vanillaAgents.default ?? vanillaAgents;
+const provider2Agent = {
+    openai: {
+        agentName: "openAIAgent",
+        defaultModel: "gpt-4o-mini",
+        keyName: "OPENAI_API_KEY",
+        maxTokens: 4096,
+    },
+    anthropic: {
+        agentName: "anthropicAgent",
+        defaultModel: "claude-sonnet-4-20250514",
+        keyName: "ANTHROPIC_API_KEY",
+        maxTokens: 4096,
+    },
+    groq: {
+        agentName: "groqAgent",
+        defaultModel: "llama-3.1-8b-instant",
+        keyName: "GROQ_API_KEY",
+        maxTokens: 4096,
+    },
+    gemini: {
+        agentName: "geminiAgent",
+        defaultModel: "gemini-2.0-flash",
+        keyName: "GEMINI_API_KEY",
+        maxTokens: 4096,
+    },
+};
+/**
+ * Get provider configuration
+ */
+export const getProviderConfig = (provider) => {
+    return provider2Agent[provider];
+};
+/**
+ * Get API key for provider
+ */
+export const getProviderApiKey = (provider) => {
+    const config = provider2Agent[provider];
+    return process.env[config.keyName];
+};
+/**
+ * Create GraphAI graph for LLM call
+ */
+const createLLMGraph = (agentName) => ({
+    version: 0.5,
+    nodes: {
+        systemPrompt: {},
+        userPrompt: {},
+        model: {},
+        temperature: {},
+        maxTokens: {},
+        llmCall: {
+            agent: agentName,
+            inputs: {
+                system: ":systemPrompt",
+                prompt: ":userPrompt",
+                model: ":model",
+                temperature: ":temperature",
+                max_tokens: ":maxTokens",
+            },
+        },
+        result: {
+            isResult: true,
+            agent: "copyAgent",
+            inputs: {
+                text: ":llmCall.text",
+            },
+        },
+    },
+});
+/**
+ * Filter script based on options (section, tags)
+ */
+export const filterScript = (script, options) => {
+    const afterSection = options.section ? filterBySection(script, options.section) : script;
+    const afterTags = options.tags && options.tags.length > 0 ? filterByTags(afterSection, options.tags) : afterSection;
+    return afterTags;
+};
+/**
+ * Get language name from code
+ */
+export const getLanguageName = (langCode) => {
+    const langMap = {
+        ja: "Japanese",
+        en: "English",
+        zh: "Chinese",
+        ko: "Korean",
+        fr: "French",
+        de: "German",
+        es: "Spanish",
+        it: "Italian",
+        pt: "Portuguese",
+        ru: "Russian",
+    };
+    return langMap[langCode] || langCode;
+};
+/**
+ * Execute LLM call with GraphAI
+ */
+export const executeLLM = async (systemPrompt, userPrompt, options, verboseMessage) => {
+    const providerConfig = getProviderConfig(options.provider);
+    const apiKey = getProviderApiKey(options.provider);
+    if (!apiKey) {
+        throw new Error(`API key not found for provider "${options.provider}". Please set the ${providerConfig.keyName} environment variable.`);
+    }
+    // Build GraphAI config
+    const config = {
+        openAIAgent: { apiKey: process.env.OPENAI_API_KEY },
+        anthropicAgent: { apiKey: process.env.ANTHROPIC_API_KEY },
+        groqAgent: { apiKey: process.env.GROQ_API_KEY },
+        geminiAgent: { apiKey: process.env.GEMINI_API_KEY },
+    };
+    // Create GraphAI instance
+    const graph = new GraphAI(createLLMGraph(providerConfig.agentName), {
+        ...agents,
+        openAIAgent,
+        anthropicAgent,
+        groqAgent,
+        geminiAgent,
+    }, { config });
+    if (options.verbose && verboseMessage) {
+        GraphAILogger.info(verboseMessage);
+    }
+    // Inject values
+    graph.injectValue("systemPrompt", systemPrompt);
+    graph.injectValue("userPrompt", userPrompt);
+    graph.injectValue("model", options.model ?? providerConfig.defaultModel);
+    graph.injectValue("temperature", options.temperature ?? 0.7);
+    graph.injectValue("maxTokens", options.maxTokens ?? providerConfig.maxTokens ?? 2048);
+    // Run graph
+    const graphResult = await graph.run();
+    // Extract text from result node
+    const resultNode = graphResult.result;
+    return resultNode?.text || "";
+};

package/lib/core/preprocessing/filter.d.ts

@@ -0,0 +1,14 @@
+import type { MulmoScript } from "mulmocast";
+import type { ExtendedScript } from "../../types/index.js";
+/**
+ * Filter beats by section (preserves meta for chaining)
+ */
+export declare const filterBySection: (script: ExtendedScript, section: string) => ExtendedScript;
+/**
+ * Filter beats by tags (preserves meta for chaining)
+ */
+export declare const filterByTags: (script: ExtendedScript, tags: string[]) => ExtendedScript;
+/**
+ * Strip variants and meta fields, converting to standard MulmoScript
+ */
+export declare const stripExtendedFields: (script: ExtendedScript) => MulmoScript;