voyageai-cli 1.15.0 → 1.18.0

package/src/lib/chunker.js

@@ -0,0 +1,341 @@
+'use strict';
+
+/**
+ * Available chunking strategies.
+ */
+const STRATEGIES = ['fixed', 'sentence', 'paragraph', 'recursive', 'markdown'];
+
+/**
+ * Default chunk options.
+ */
+const DEFAULTS = {
+  size: 512,
+  overlap: 50,
+  minSize: 20,
+};
+
+// ── Sentence splitting ──
+
+/**
+ * Split text into sentences. Handles common abbreviations and edge cases.
+ * @param {string} text
+ * @returns {string[]}
+ */
+function splitSentences(text) {
+  // Split on sentence-ending punctuation followed by whitespace or EOL.
+  // Negative lookbehind for common abbreviations (Mr., Dr., etc.)
+  const parts = text.split(/(?<=[.!?])\s+(?=[A-Z\u00C0-\u024F"])/);
+  return parts.map(s => s.trim()).filter(s => s.length > 0);
+}
+
+// ── Strategy implementations ──
+
+/**
+ * Fixed-size chunking with character count and overlap.
+ * @param {string} text
+ * @param {object} opts
+ * @param {number} opts.size - Target chunk size in characters
+ * @param {number} opts.overlap - Overlap between chunks in characters
+ * @returns {string[]}
+ */
+function chunkFixed(text, opts) {
+  const { size, overlap } = opts;
+  const chunks = [];
+  let start = 0;
+
+  while (start < text.length) {
+    const end = start + size;
+    chunks.push(text.slice(start, end).trim());
+    start = end - overlap;
+    if (start >= text.length) break;
+    // Prevent infinite loop with tiny overlap
+    if (end >= text.length) break;
+  }
+
+  return chunks.filter(c => c.length >= (opts.minSize || DEFAULTS.minSize));
+}
+
+/**
+ * Sentence-boundary chunking. Groups sentences until size limit.
+ * @param {string} text
+ * @param {object} opts
+ * @returns {string[]}
+ */
+function chunkSentence(text, opts) {
+  const { size, overlap } = opts;
+  const sentences = splitSentences(text);
+  return groupUnits(sentences, size, overlap, opts.minSize || DEFAULTS.minSize);
+}
+
+/**
+ * Paragraph chunking. Splits on double newlines, groups if needed.
+ * @param {string} text
+ * @param {object} opts
+ * @returns {string[]}
+ */
+function chunkParagraph(text, opts) {
+  const { size, overlap } = opts;
+  const paragraphs = text.split(/\n\s*\n/).map(p => p.trim()).filter(p => p.length > 0);
+  return groupUnits(paragraphs, size, overlap, opts.minSize || DEFAULTS.minSize);
+}
+
+/**
+ * Recursive chunking. Tries largest delimiters first, falls back to smaller.
+ * This is the most commonly used strategy for RAG pipelines.
+ * @param {string} text
+ * @param {object} opts
+ * @returns {string[]}
+ */
+function chunkRecursive(text, opts) {
+  const { size, minSize } = opts;
+  const separators = ['\n\n', '\n', '. ', '! ', '? ', '; ', ', ', ' '];
+
+  return recursiveSplit(text, separators, size, minSize || DEFAULTS.minSize);
+}
+
+/**
+ * Internal recursive split implementation.
+ * @param {string} text
+ * @param {string[]} separators
+ * @param {number} maxSize
+ * @param {number} minSize
+ * @returns {string[]}
+ */
+function recursiveSplit(text, separators, maxSize, minSize) {
+  if (text.length <= maxSize) {
+    return text.trim().length >= minSize ? [text.trim()] : [];
+  }
+
+  // Find the first separator that exists in the text
+  let sep = null;
+  for (const s of separators) {
+    if (text.includes(s)) {
+      sep = s;
+      break;
+    }
+  }
+
+  // If no separator found, hard-split by characters
+  if (sep === null) {
+    const chunks = [];
+    for (let i = 0; i < text.length; i += maxSize) {
+      const chunk = text.slice(i, i + maxSize).trim();
+      if (chunk.length >= minSize) chunks.push(chunk);
+    }
+    return chunks;
+  }
+
+  // Split on this separator and greedily merge pieces under maxSize
+  const parts = text.split(sep);
+  const chunks = [];
+  let current = '';
+
+  for (const part of parts) {
+    const candidate = current ? current + sep + part : part;
+
+    if (candidate.length <= maxSize) {
+      current = candidate;
+    } else {
+      // Flush current chunk
+      if (current.trim().length >= minSize) {
+        chunks.push(current.trim());
+      }
+      // If this single part exceeds maxSize, recurse with next separator level
+      if (part.length > maxSize) {
+        const remainingSeps = separators.slice(separators.indexOf(sep) + 1);
+        chunks.push(...recursiveSplit(part, remainingSeps, maxSize, minSize));
+        current = '';
+      } else {
+        current = part;
+      }
+    }
+  }
+
+  // Flush remainder
+  if (current.trim().length >= minSize) {
+    chunks.push(current.trim());
+  }
+
+  return chunks;
+}
+
+/**
+ * Markdown-aware chunking. Splits on headings, preserves structure.
+ * Each heading starts a new chunk; content under it is grouped.
+ * @param {string} text
+ * @param {object} opts
+ * @returns {string[]}
+ */
+function chunkMarkdown(text, opts) {
+  const { size, minSize } = opts;
+
+  // Split on markdown headings (# through ######)
+  const headingPattern = /^(#{1,6}\s.+)$/gm;
+  const sections = [];
+  let lastIndex = 0;
+  let match;
+
+  while ((match = headingPattern.exec(text)) !== null) {
+    // Content before this heading
+    if (match.index > lastIndex) {
+      const content = text.slice(lastIndex, match.index).trim();
+      if (content) {
+        if (sections.length > 0) {
+          // Append to previous section
+          sections[sections.length - 1].content += '\n\n' + content;
+        } else {
+          sections.push({ heading: '', content });
+        }
+      }
+    }
+    sections.push({ heading: match[1], content: '' });
+    lastIndex = match.index + match[0].length;
+  }
+
+  // Remaining content after last heading
+  if (lastIndex < text.length) {
+    const content = text.slice(lastIndex).trim();
+    if (content) {
+      if (sections.length > 0) {
+        sections[sections.length - 1].content += '\n\n' + content;
+      } else {
+        sections.push({ heading: '', content });
+      }
+    }
+  }
+
+  // Build chunks from sections, splitting large sections recursively
+  const chunks = [];
+  for (const section of sections) {
+    const full = section.heading
+      ? section.heading + '\n\n' + section.content.trim()
+      : section.content.trim();
+
+    if (!full || full.length < (minSize || DEFAULTS.minSize)) continue;
+
+    if (full.length <= size) {
+      chunks.push(full);
+    } else {
+      // Section too large — recursively split the content, prepend heading to first chunk
+      const subChunks = chunkRecursive(section.content.trim(), opts);
+      for (let i = 0; i < subChunks.length; i++) {
+        if (i === 0 && section.heading) {
+          chunks.push(section.heading + '\n\n' + subChunks[i]);
+        } else {
+          chunks.push(subChunks[i]);
+        }
+      }
+    }
+  }
+
+  return chunks;
+}
+
+// ── Shared helpers ──
+
+/**
+ * Group text units (sentences, paragraphs) into chunks under a size limit.
+ * Supports overlap by re-including trailing units from the previous chunk.
+ * @param {string[]} units
+ * @param {number} maxSize
+ * @param {number} overlapChars
+ * @param {number} minSize
+ * @returns {string[]}
+ */
+function groupUnits(units, maxSize, overlapChars, minSize) {
+  const chunks = [];
+  let current = [];
+  let currentLen = 0;
+
+  for (const unit of units) {
+    const addLen = current.length > 0 ? unit.length + 1 : unit.length; // +1 for space
+
+    if (currentLen + addLen > maxSize && current.length > 0) {
+      chunks.push(current.join(' ').trim());
+
+      // Overlap: keep trailing units that fit within overlap budget
+      if (overlapChars > 0) {
+        let overlapUnits = [];
+        let overlapLen = 0;
+        for (let i = current.length - 1; i >= 0; i--) {
+          if (overlapLen + current[i].length + 1 > overlapChars) break;
+          overlapUnits.unshift(current[i]);
+          overlapLen += current[i].length + 1;
+        }
+        current = overlapUnits;
+        currentLen = overlapLen;
+      } else {
+        current = [];
+        currentLen = 0;
+      }
+    }
+
+    current.push(unit);
+    currentLen += addLen;
+  }
+
+  // Flush remainder
+  if (current.length > 0) {
+    const text = current.join(' ').trim();
+    if (text.length >= minSize) chunks.push(text);
+  }
+
+  return chunks;
+}
+
+// ── Token estimation ──
+
+/**
+ * Rough token estimate. ~4 chars per token for English text.
+ * @param {string} text
+ * @returns {number}
+ */
+function estimateTokens(text) {
+  return Math.ceil(text.length / 4);
+}
+
+// ── Public API ──
+
+/**
+ * Chunk text using the specified strategy.
+ * @param {string} text - Input text
+ * @param {object} [options]
+ * @param {string} [options.strategy='recursive'] - Chunking strategy
+ * @param {number} [options.size=512] - Target chunk size in characters
+ * @param {number} [options.overlap=50] - Overlap between chunks in characters
+ * @param {number} [options.minSize=20] - Minimum chunk size
+ * @returns {string[]} Array of text chunks
+ */
+function chunk(text, options = {}) {
+  const opts = {
+    strategy: options.strategy || 'recursive',
+    size: options.size || DEFAULTS.size,
+    overlap: options.overlap != null ? options.overlap : DEFAULTS.overlap,
+    minSize: options.minSize || DEFAULTS.minSize,
+  };
+
+  if (!text || text.trim().length === 0) return [];
+
+  switch (opts.strategy) {
+    case 'fixed':
+      return chunkFixed(text, opts);
+    case 'sentence':
+      return chunkSentence(text, opts);
+    case 'paragraph':
+      return chunkParagraph(text, opts);
+    case 'recursive':
+      return chunkRecursive(text, opts);
+    case 'markdown':
+      return chunkMarkdown(text, opts);
+    default:
+      throw new Error(`Unknown chunking strategy: ${opts.strategy}. Available: ${STRATEGIES.join(', ')}`);
+  }
+}
+
+module.exports = {
+  chunk,
+  splitSentences,
+  estimateTokens,
+  STRATEGIES,
+  DEFAULTS,
+};

package/src/lib/project.js

@@ -0,0 +1,122 @@
+'use strict';
+
+const fs = require('fs');
+const path = require('path');
+
+const PROJECT_FILE = '.vai.json';
+const PROJECT_VERSION = 1;
+
+/**
+ * Search for .vai.json starting from startDir, walking up to root.
+ * @param {string} [startDir] - Directory to start from (default: cwd)
+ * @returns {string|null} Absolute path to .vai.json or null
+ */
+function findProjectFile(startDir) {
+  let dir = path.resolve(startDir || process.cwd());
+  const root = path.parse(dir).root;
+
+  while (dir !== root) {
+    const candidate = path.join(dir, PROJECT_FILE);
+    if (fs.existsSync(candidate)) return candidate;
+    dir = path.dirname(dir);
+  }
+  // Check root too
+  const rootCandidate = path.join(root, PROJECT_FILE);
+  if (fs.existsSync(rootCandidate)) return rootCandidate;
+
+  return null;
+}
+
+/**
+ * Load project config from .vai.json.
+ * @param {string} [startDir] - Directory to start searching from
+ * @returns {{ config: object, filePath: string|null }}
+ */
+function loadProject(startDir) {
+  const filePath = findProjectFile(startDir);
+  if (!filePath) return { config: {}, filePath: null };
+
+  try {
+    const raw = fs.readFileSync(filePath, 'utf-8');
+    return { config: JSON.parse(raw), filePath };
+  } catch (err) {
+    return { config: {}, filePath };
+  }
+}
+
+/**
+ * Save project config to .vai.json.
+ * @param {object} config - Project configuration
+ * @param {string} [targetPath] - Path to write (default: cwd/.vai.json)
+ */
+function saveProject(config, targetPath) {
+  const filePath = targetPath || path.join(process.cwd(), PROJECT_FILE);
+  const output = { version: PROJECT_VERSION, ...config };
+  fs.writeFileSync(filePath, JSON.stringify(output, null, 2) + '\n', 'utf-8');
+  return filePath;
+}
+
+/**
+ * Merge project config with CLI options. CLI options take precedence.
+ * Only merges known keys — doesn't blindly spread everything.
+ * @param {object} projectConfig - From .vai.json
+ * @param {object} cliOpts - From commander
+ * @returns {object} Merged options
+ */
+function mergeOptions(projectConfig, cliOpts) {
+  const merged = {};
+
+  // Map of project config keys → CLI option keys
+  const keys = [
+    'model', 'db', 'collection', 'field', 'inputType',
+    'dimensions', 'index',
+  ];
+
+  for (const key of keys) {
+    // CLI explicit value wins, then project config, then undefined
+    if (cliOpts[key] !== undefined) {
+      merged[key] = cliOpts[key];
+    } else if (projectConfig[key] !== undefined) {
+      merged[key] = projectConfig[key];
+    }
+  }
+
+  // Chunk config nests under project.chunk
+  if (projectConfig.chunk) {
+    merged.chunk = { ...projectConfig.chunk };
+  }
+
+  return merged;
+}
+
+/**
+ * Default project config scaffold.
+ * @returns {object}
+ */
+function defaultProjectConfig() {
+  return {
+    version: PROJECT_VERSION,
+    model: 'voyage-4-large',
+    db: '',
+    collection: '',
+    field: 'embedding',
+    inputType: 'document',
+    dimensions: 1024,
+    index: 'vector_index',
+    chunk: {
+      strategy: 'recursive',
+      size: 512,
+      overlap: 50,
+    },
+  };
+}
+
+module.exports = {
+  PROJECT_FILE,
+  PROJECT_VERSION,
+  findProjectFile,
+  loadProject,
+  saveProject,
+  mergeOptions,
+  defaultProjectConfig,
+};

package/src/lib/readers.js

@@ -0,0 +1,239 @@
+'use strict';
+
+const fs = require('fs');
+const path = require('path');
+
+/**
+ * Supported file extensions and their reader types.
+ */
+const SUPPORTED_EXTENSIONS = {
+  '.txt': 'text',
+  '.md': 'text',
+  '.markdown': 'text',
+  '.rst': 'text',
+  '.html': 'html',
+  '.htm': 'html',
+  '.json': 'json',
+  '.jsonl': 'jsonl',
+  '.ndjson': 'jsonl',
+  '.csv': 'text',
+  '.pdf': 'pdf',
+};
+
+/**
+ * Check if a file extension is supported.
+ * @param {string} filePath
+ * @returns {boolean}
+ */
+function isSupported(filePath) {
+  const ext = path.extname(filePath).toLowerCase();
+  return ext in SUPPORTED_EXTENSIONS;
+}
+
+/**
+ * Get the reader type for a file.
+ * @param {string} filePath
+ * @returns {string|null}
+ */
+function getReaderType(filePath) {
+  const ext = path.extname(filePath).toLowerCase();
+  return SUPPORTED_EXTENSIONS[ext] || null;
+}
+
+/**
+ * Read a text file (txt, md, rst, csv).
+ * @param {string} filePath
+ * @returns {Promise<string>}
+ */
+async function readTextFile(filePath) {
+  return fs.readFileSync(filePath, 'utf-8');
+}
+
+/**
+ * Read an HTML file and strip tags to plain text.
+ * Lightweight — no external dependencies.
+ * @param {string} filePath
+ * @returns {Promise<string>}
+ */
+async function readHtmlFile(filePath) {
+  const html = fs.readFileSync(filePath, 'utf-8');
+  return stripHtml(html);
+}
+
+/**
+ * Strip HTML tags and decode common entities.
+ * @param {string} html
+ * @returns {string}
+ */
+function stripHtml(html) {
+  return html
+    // Remove script and style blocks
+    .replace(/<script[\s\S]*?<\/script>/gi, '')
+    .replace(/<style[\s\S]*?<\/style>/gi, '')
+    // Replace block elements with newlines
+    .replace(/<\/?(p|div|br|h[1-6]|li|tr|blockquote|section|article|header|footer|nav|pre)[^>]*>/gi, '\n')
+    // Remove remaining tags
+    .replace(/<[^>]+>/g, '')
+    // Decode common entities
+    .replace(/&amp;/g, '&')
+    .replace(/&lt;/g, '<')
+    .replace(/&gt;/g, '>')
+    .replace(/&quot;/g, '"')
+    .replace(/&#39;/g, "'")
+    .replace(/&nbsp;/g, ' ')
+    // Collapse whitespace
+    .replace(/\n{3,}/g, '\n\n')
+    .trim();
+}
+
+/**
+ * Read a JSON file. Extracts text from objects using a text field.
+ * Supports JSON array of objects or a single object with a text field.
+ * @param {string} filePath
+ * @param {string} [textField='text'] - Field name containing text
+ * @returns {Promise<Array<{text: string, metadata: object}>>}
+ */
+async function readJsonFile(filePath, textField = 'text') {
+  const raw = fs.readFileSync(filePath, 'utf-8');
+  const data = JSON.parse(raw);
+
+  if (Array.isArray(data)) {
+    return data.map((item, i) => {
+      const text = item[textField];
+      if (!text) throw new Error(`Missing "${textField}" field in array item ${i}`);
+      const metadata = { ...item };
+      delete metadata[textField];
+      return { text, metadata };
+    });
+  }
+
+  if (typeof data === 'object' && data[textField]) {
+    const metadata = { ...data };
+    delete metadata[textField];
+    return [{ text: data[textField], metadata }];
+  }
+
+  throw new Error(`JSON file must be an array of objects or an object with a "${textField}" field`);
+}
+
+/**
+ * Read a JSONL/NDJSON file.
+ * @param {string} filePath
+ * @param {string} [textField='text']
+ * @returns {Promise<Array<{text: string, metadata: object}>>}
+ */
+async function readJsonlFile(filePath, textField = 'text') {
+  const raw = fs.readFileSync(filePath, 'utf-8');
+  const lines = raw.split('\n').filter(l => l.trim().length > 0);
+
+  return lines.map((line, i) => {
+    const item = JSON.parse(line);
+    const text = item[textField];
+    if (!text) throw new Error(`Missing "${textField}" field on line ${i + 1}`);
+    const metadata = { ...item };
+    delete metadata[textField];
+    return { text, metadata };
+  });
+}
+
+/**
+ * Read a PDF file. Requires optional `pdf-parse` dependency.
+ * @param {string} filePath
+ * @returns {Promise<string>}
+ */
+async function readPdfFile(filePath) {
+  let pdfParse;
+  try {
+    pdfParse = require('pdf-parse');
+  } catch {
+    throw new Error(
+      'PDF support requires the "pdf-parse" package.\n' +
+      'Install it: npm install pdf-parse\n' +
+      'Then retry your command.'
+    );
+  }
+  const buffer = fs.readFileSync(filePath);
+  const data = await pdfParse(buffer);
+  return data.text;
+}
+
+/**
+ * Read a single file and return its text content.
+ * For structured files (JSON/JSONL), returns array of {text, metadata}.
+ * For text files, returns the raw text string.
+ * @param {string} filePath
+ * @param {object} [opts]
+ * @param {string} [opts.textField='text'] - Field name for JSON/JSONL
+ * @returns {Promise<string|Array<{text: string, metadata: object}>>}
+ */
+async function readFile(filePath, opts = {}) {
+  const type = getReaderType(filePath);
+  if (!type) {
+    throw new Error(`Unsupported file type: ${path.extname(filePath)}. Supported: ${Object.keys(SUPPORTED_EXTENSIONS).join(', ')}`);
+  }
+
+  switch (type) {
+    case 'text':
+      return readTextFile(filePath);
+    case 'html':
+      return readHtmlFile(filePath);
+    case 'json':
+      return readJsonFile(filePath, opts.textField || 'text');
+    case 'jsonl':
+      return readJsonlFile(filePath, opts.textField || 'text');
+    case 'pdf':
+      return readPdfFile(filePath);
+    default:
+      throw new Error(`No reader for type: ${type}`);
+  }
+}
+
+/**
+ * Recursively scan a directory for supported files.
+ * @param {string} dirPath
+ * @param {object} [opts]
+ * @param {string[]} [opts.extensions] - Filter to specific extensions
+ * @param {string[]} [opts.ignore] - Directory names to skip
+ * @returns {string[]} Array of absolute file paths
+ */
+function scanDirectory(dirPath, opts = {}) {
+  const ignore = new Set(opts.ignore || ['node_modules', '.git', '.vai', '__pycache__', '.DS_Store']);
+  const extensions = opts.extensions
+    ? new Set(opts.extensions.map(e => e.startsWith('.') ? e : '.' + e))
+    : null;
+
+  const results = [];
+
+  function walk(dir) {
+    const entries = fs.readdirSync(dir, { withFileTypes: true });
+    for (const entry of entries) {
+      if (entry.name.startsWith('.') && ignore.has(entry.name)) continue;
+      if (ignore.has(entry.name)) continue;
+
+      const fullPath = path.join(dir, entry.name);
+
+      if (entry.isDirectory()) {
+        walk(fullPath);
+      } else if (entry.isFile()) {
+        const ext = path.extname(entry.name).toLowerCase();
+        if (extensions) {
+          if (extensions.has(ext)) results.push(fullPath);
+        } else if (SUPPORTED_EXTENSIONS[ext]) {
+          results.push(fullPath);
+        }
+      }
+    }
+  }
+
+  walk(path.resolve(dirPath));
+  return results.sort();
+}
+
+module.exports = {
+  SUPPORTED_EXTENSIONS,
+  isSupported,
+  getReaderType,
+  readFile,
+  scanDirectory,
+  stripHtml,
+};