llmist 2.3.0 → 2.5.0

package/dist/cli.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import "./chunk-ZDNV7DDO.js";
+import "./chunk-IHSZUAYN.js";
 import {
   AgentBuilder,
   BaseGadget,
@@ -10,14 +10,20 @@ import {
   LLMMessageBuilder,
   LLMist,
   MODEL_ALIASES,
+  audioFromBuffer,
   createGadget,
   createLogger,
+  detectAudioMimeType,
+  detectImageMimeType,
+  extractText,
+  imageFromBuffer,
   init_builder,
   init_client,
   init_constants,
   init_create_gadget,
   init_exceptions,
   init_gadget,
+  init_input_content,
   init_logger,
   init_messages,
   init_model_shortcuts,
@@ -26,8 +32,9 @@ import {
   init_schema_validator,
   resolveModel,
   schemaToJSONSchema,
+  text,
   validateGadgetSchema
-} from "./chunk-GANXNBIZ.js";
+} from "./chunk-YHS2DYXP.js";
 
 // src/cli/constants.ts
 var CLI_NAME = "llmist";
@@ -36,7 +43,10 @@ var COMMANDS = {
   complete: "complete",
   agent: "agent",
   models: "models",
-  gadget: "gadget"
+  gadget: "gadget",
+  image: "image",
+  speech: "speech",
+  vision: "vision"
 };
 var LOG_LEVELS = ["silly", "trace", "debug", "info", "warn", "error", "fatal"];
 var DEFAULT_MODEL = "openai:gpt-5-nano";
@@ -57,7 +67,20 @@ var OPTION_FLAGS = {
   docker: "--docker",
   dockerRo: "--docker-ro",
   noDocker: "--no-docker",
-  dockerDev: "--docker-dev"
+  dockerDev: "--docker-dev",
+  // Multimodal input options
+  inputImage: "--image <path>",
+  inputAudio: "--audio <path>",
+  // Image generation options
+  imageSize: "--size <size>",
+  imageQuality: "--quality <quality>",
+  imageCount: "-n, --count <number>",
+  imageOutput: "-o, --output <path>",
+  // Speech generation options
+  voice: "--voice <name>",
+  speechFormat: "--format <format>",
+  speechSpeed: "--speed <value>",
+  speechOutput: "-o, --output <path>"
 };
 var OPTION_DESCRIPTIONS = {
   model: "Model identifier, e.g. openai:gpt-5-nano or anthropic:claude-sonnet-4-5.",
@@ -73,10 +96,23 @@ var OPTION_DESCRIPTIONS = {
   noBuiltins: "Disable built-in gadgets (AskUser, TellUser).",
   noBuiltinInteraction: "Disable interactive gadgets (AskUser) while keeping TellUser.",
   quiet: "Suppress all output except content (text and TellUser messages).",
+  // Multimodal input descriptions
+  inputImage: "Image file to include with the prompt (vision models).",
+  inputAudio: "Audio file to include with the prompt (Gemini only).",
   docker: "Run agent in a Docker sandbox container for security isolation.",
   dockerRo: "Run in Docker with current directory mounted read-only.",
   noDocker: "Disable Docker sandboxing (override config).",
-  dockerDev: "Run in Docker dev mode (mount local source instead of npm install)."
+  dockerDev: "Run in Docker dev mode (mount local source instead of npm install).",
+  // Image generation descriptions
+  imageSize: "Image size/aspect ratio, e.g. '1024x1024', '1:1', '16:9'.",
+  imageQuality: "Image quality: 'standard', 'hd', 'low', 'medium', 'high'.",
+  imageCount: "Number of images to generate (model dependent, usually 1-4).",
+  imageOutput: "Output path for the generated image. Defaults to stdout if not specified.",
+  // Speech generation descriptions
+  voice: "Voice name for speech generation, e.g. 'nova', 'alloy', 'Zephyr'.",
+  speechFormat: "Audio format: 'mp3', 'opus', 'aac', 'flac', 'wav', 'pcm'.",
+  speechSpeed: "Speech speed multiplier (0.25 to 4.0, default 1.0).",
+  speechOutput: "Output path for audio file. Defaults to stdout if not specified."
 };
 var SUMMARY_PREFIX = "[llmist]";
 
@@ -86,7 +122,7 @@ import { Command, InvalidArgumentError as InvalidArgumentError2 } from "commande
 // package.json
 var package_default = {
   name: "llmist",
-  version: "2.2.0",
+  version: "2.5.0",
   description: "TypeScript LLM client with streaming tool execution. Tools fire mid-stream. Built-in function calling works with any model\u2014no structured outputs or native tool support required.",
   type: "module",
   main: "dist/index.cjs",
@@ -226,6 +262,7 @@ function isAbortError(error) {
 }
 
 // src/cli/agent-command.ts
+init_input_content();
 init_registry();
 init_constants();
 
@@ -550,6 +587,75 @@ var finish = createGadget({
 });
 var builtinGadgets = [askUser, tellUser, finish];
 
+// src/cli/file-utils.ts
+init_input_content();
+import { readFile, stat } from "node:fs/promises";
+import { resolve as resolve2 } from "node:path";
+var DEFAULT_MAX_FILE_SIZE = 50 * 1024 * 1024;
+function formatFileSize(bytes) {
+  if (bytes < 1024) return `${bytes} bytes`;
+  if (bytes < 1024 * 1024) return `${(bytes / 1024).toFixed(1)} KB`;
+  if (bytes < 1024 * 1024 * 1024) return `${(bytes / (1024 * 1024)).toFixed(1)} MB`;
+  return `${(bytes / (1024 * 1024 * 1024)).toFixed(1)} GB`;
+}
+async function checkFileSize(absolutePath, filePath, maxSize) {
+  const stats = await stat(absolutePath);
+  if (stats.size > maxSize) {
+    throw new Error(
+      `File "${filePath}" is too large (${formatFileSize(stats.size)}). Maximum allowed size is ${formatFileSize(maxSize)}. Consider compressing the file or using a smaller version.`
+    );
+  }
+}
+async function readImageFile(filePath, options = {}) {
+  const absolutePath = resolve2(filePath);
+  const maxFileSize = options.maxFileSize ?? DEFAULT_MAX_FILE_SIZE;
+  let buffer;
+  try {
+    await checkFileSize(absolutePath, filePath, maxFileSize);
+    buffer = await readFile(absolutePath);
+  } catch (error) {
+    const message = error instanceof Error ? error.message : String(error);
+    throw new Error(`Failed to read image file "${filePath}": ${message}`);
+  }
+  const mimeType = detectImageMimeType(buffer);
+  if (!mimeType) {
+    throw new Error(
+      `File "${filePath}" is not a supported image format. Supported formats: JPEG, PNG, GIF, WebP`
+    );
+  }
+  return imageFromBuffer(buffer, mimeType);
+}
+async function readAudioFile(filePath, options = {}) {
+  const absolutePath = resolve2(filePath);
+  const maxFileSize = options.maxFileSize ?? DEFAULT_MAX_FILE_SIZE;
+  let buffer;
+  try {
+    await checkFileSize(absolutePath, filePath, maxFileSize);
+    buffer = await readFile(absolutePath);
+  } catch (error) {
+    const message = error instanceof Error ? error.message : String(error);
+    throw new Error(`Failed to read audio file "${filePath}": ${message}`);
+  }
+  const mimeType = detectAudioMimeType(buffer);
+  if (!mimeType) {
+    throw new Error(
+      `File "${filePath}" is not a supported audio format. Supported formats: MP3, WAV, OGG, WebM`
+    );
+  }
+  return audioFromBuffer(buffer, mimeType);
+}
+async function readFileBuffer(filePath, options = {}) {
+  const absolutePath = resolve2(filePath);
+  const maxFileSize = options.maxFileSize ?? DEFAULT_MAX_FILE_SIZE;
+  try {
+    await checkFileSize(absolutePath, filePath, maxFileSize);
+    return await readFile(absolutePath);
+  } catch (error) {
+    const message = error instanceof Error ? error.message : String(error);
+    throw new Error(`Failed to read file "${filePath}": ${message}`);
+  }
+}
+
 // src/cli/gadgets.ts
 init_gadget();
 import fs5 from "node:fs";
@@ -713,7 +819,7 @@ ${formattedList}`;
 // src/cli/builtins/filesystem/read-file.ts
 import fs3 from "node:fs";
 import { z as z3 } from "zod";
-var readFile = createGadget({
+var readFile2 = createGadget({
   name: "ReadFile",
   description: "Read the entire content of a file and return it as text. The file path must be within the current working directory or its subdirectories.",
   schema: z3.object({
@@ -984,7 +1090,7 @@ error: ${message}`;
 // src/cli/builtins/index.ts
 var builtinGadgetRegistry = {
   ListDirectory: listDirectory,
-  ReadFile: readFile,
+  ReadFile: readFile2,
   WriteFile: writeFile,
   EditFile: editFile,
   RunCommand: runCommand
@@ -1119,6 +1225,7 @@ async function loadGadgets(specifiers, cwd, importer = (specifier) => import(spe
 }
 
 // src/cli/llm-logging.ts
+init_messages();
 import { mkdir, writeFile as writeFile2 } from "node:fs/promises";
 import { homedir } from "node:os";
 import { join } from "node:path";
@@ -1136,7 +1243,7 @@ function formatLlmRequest(messages) {
   const lines = [];
   for (const msg of messages) {
     lines.push(`=== ${msg.role.toUpperCase()} ===`);
-    lines.push(msg.content ?? "");
+    lines.push(msg.content ? extractText(msg.content) : "");
     lines.push("");
   }
   return lines.join("\n");
@@ -1210,9 +1317,9 @@ function ensureMarkedConfigured() {
     markedConfigured = true;
   }
 }
-function renderMarkdown(text) {
+function renderMarkdown(text2) {
   ensureMarkedConfigured();
-  let rendered = marked.parse(text);
+  let rendered = marked.parse(text2);
   rendered = rendered.replace(/\*\*(.+?)\*\*/g, (_, content) => chalk3.bold(content)).replace(/(?<!\*)\*(\S[^*]*)\*(?!\*)/g, (_, content) => chalk3.italic(content));
   return rendered.trimEnd();
 }
@@ -1226,8 +1333,8 @@ function createRainbowSeparator() {
   }
   return result;
 }
-function renderMarkdownWithSeparators(text) {
-  const rendered = renderMarkdown(text);
+function renderMarkdownWithSeparators(text2) {
+  const rendered = renderMarkdown(text2);
   const separator = createRainbowSeparator();
   return `
 ${separator}
@@ -1395,12 +1502,12 @@ var StreamPrinter = class {
    *
    * @param text - Text to write
    */
-  write(text) {
-    if (!text) {
+  write(text2) {
+    if (!text2) {
       return;
     }
-    this.target.write(text);
-    this.endedWithNewline = text.endsWith("\n");
+    this.target.write(text2);
+    this.endedWithNewline = text2.endsWith("\n");
   }
   /**
    * Ensures output ends with a newline by writing one if needed.
@@ -1879,7 +1986,7 @@ function addCompleteOptions(cmd, defaults) {
     OPTION_DESCRIPTIONS.maxTokens,
     createNumericParser({ label: "Max tokens", integer: true, min: 1 }),
     defaults?.["max-tokens"]
-  ).option(OPTION_FLAGS.quiet, OPTION_DESCRIPTIONS.quiet, defaults?.quiet).option(OPTION_FLAGS.logLlmRequests, OPTION_DESCRIPTIONS.logLlmRequests, defaults?.["log-llm-requests"]);
+  ).option(OPTION_FLAGS.quiet, OPTION_DESCRIPTIONS.quiet, defaults?.quiet).option(OPTION_FLAGS.logLlmRequests, OPTION_DESCRIPTIONS.logLlmRequests, defaults?.["log-llm-requests"]).option(OPTION_FLAGS.inputImage, OPTION_DESCRIPTIONS.inputImage).option(OPTION_FLAGS.inputAudio, OPTION_DESCRIPTIONS.inputAudio);
 }
 function addAgentOptions(cmd, defaults) {
   const gadgetAccumulator = (value, previous = []) => [
@@ -1903,7 +2010,7 @@ function addAgentOptions(cmd, defaults) {
     OPTION_FLAGS.noBuiltinInteraction,
     OPTION_DESCRIPTIONS.noBuiltinInteraction,
     defaults?.["builtin-interaction"] !== false
-  ).option(OPTION_FLAGS.quiet, OPTION_DESCRIPTIONS.quiet, defaults?.quiet).option(OPTION_FLAGS.logLlmRequests, OPTION_DESCRIPTIONS.logLlmRequests, defaults?.["log-llm-requests"]).option(OPTION_FLAGS.docker, OPTION_DESCRIPTIONS.docker).option(OPTION_FLAGS.dockerRo, OPTION_DESCRIPTIONS.dockerRo).option(OPTION_FLAGS.noDocker, OPTION_DESCRIPTIONS.noDocker).option(OPTION_FLAGS.dockerDev, OPTION_DESCRIPTIONS.dockerDev);
+  ).option(OPTION_FLAGS.quiet, OPTION_DESCRIPTIONS.quiet, defaults?.quiet).option(OPTION_FLAGS.logLlmRequests, OPTION_DESCRIPTIONS.logLlmRequests, defaults?.["log-llm-requests"]).option(OPTION_FLAGS.inputImage, OPTION_DESCRIPTIONS.inputImage).option(OPTION_FLAGS.inputAudio, OPTION_DESCRIPTIONS.inputAudio).option(OPTION_FLAGS.docker, OPTION_DESCRIPTIONS.docker).option(OPTION_FLAGS.dockerRo, OPTION_DESCRIPTIONS.dockerRo).option(OPTION_FLAGS.noDocker, OPTION_DESCRIPTIONS.noDocker).option(OPTION_FLAGS.dockerDev, OPTION_DESCRIPTIONS.dockerDev);
 }
 function configToCompleteOptions(config) {
   const result = {};
@@ -2108,6 +2215,22 @@ var AGENT_CONFIG_KEYS = /* @__PURE__ */ new Set([
   "docker-cwd-permission"
   // Override CWD mount permission for this profile
 ]);
+var IMAGE_CONFIG_KEYS = /* @__PURE__ */ new Set([
+  "model",
+  "size",
+  "quality",
+  "count",
+  "output",
+  "quiet"
+]);
+var SPEECH_CONFIG_KEYS = /* @__PURE__ */ new Set([
+  "model",
+  "voice",
+  "format",
+  "speed",
+  "output",
+  "quiet"
+]);
 var CUSTOM_CONFIG_KEYS = /* @__PURE__ */ new Set([
   ...COMPLETE_CONFIG_KEYS,
   ...AGENT_CONFIG_KEYS,
@@ -2368,6 +2491,75 @@ function validateAgentConfig(raw, section) {
   }
   return result;
 }
+function validateImageConfig(raw, section) {
+  if (typeof raw !== "object" || raw === null) {
+    throw new ConfigError(`[${section}] must be a table`);
+  }
+  const rawObj = raw;
+  for (const key of Object.keys(rawObj)) {
+    if (!IMAGE_CONFIG_KEYS.has(key)) {
+      throw new ConfigError(`[${section}].${key} is not a valid option`);
+    }
+  }
+  const result = {};
+  if ("model" in rawObj) {
+    result.model = validateString(rawObj.model, "model", section);
+  }
+  if ("size" in rawObj) {
+    result.size = validateString(rawObj.size, "size", section);
+  }
+  if ("quality" in rawObj) {
+    result.quality = validateString(rawObj.quality, "quality", section);
+  }
+  if ("count" in rawObj) {
+    result.count = validateNumber(rawObj.count, "count", section, {
+      integer: true,
+      min: 1,
+      max: 10
+    });
+  }
+  if ("output" in rawObj) {
+    result.output = validateString(rawObj.output, "output", section);
+  }
+  if ("quiet" in rawObj) {
+    result.quiet = validateBoolean(rawObj.quiet, "quiet", section);
+  }
+  return result;
+}
+function validateSpeechConfig(raw, section) {
+  if (typeof raw !== "object" || raw === null) {
+    throw new ConfigError(`[${section}] must be a table`);
+  }
+  const rawObj = raw;
+  for (const key of Object.keys(rawObj)) {
+    if (!SPEECH_CONFIG_KEYS.has(key)) {
+      throw new ConfigError(`[${section}].${key} is not a valid option`);
+    }
+  }
+  const result = {};
+  if ("model" in rawObj) {
+    result.model = validateString(rawObj.model, "model", section);
+  }
+  if ("voice" in rawObj) {
+    result.voice = validateString(rawObj.voice, "voice", section);
+  }
+  if ("format" in rawObj) {
+    result.format = validateString(rawObj.format, "format", section);
+  }
+  if ("speed" in rawObj) {
+    result.speed = validateNumber(rawObj.speed, "speed", section, {
+      min: 0.25,
+      max: 4
+    });
+  }
+  if ("output" in rawObj) {
+    result.output = validateString(rawObj.output, "output", section);
+  }
+  if ("quiet" in rawObj) {
+    result.quiet = validateBoolean(rawObj.quiet, "quiet", section);
+  }
+  return result;
+}
 function validateStringOrBoolean(value, field, section) {
   if (typeof value === "string" || typeof value === "boolean") {
     return value;
@@ -2490,6 +2682,10 @@ function validateConfig(raw, configPath) {
         result.complete = validateCompleteConfig(value, key);
       } else if (key === "agent") {
         result.agent = validateAgentConfig(value, key);
+      } else if (key === "image") {
+        result.image = validateImageConfig(value, key);
+      } else if (key === "speech") {
+        result.speech = validateSpeechConfig(value, key);
       } else if (key === "prompts") {
         result.prompts = validatePromptsConfig(value, key);
       } else if (key === "docker") {
@@ -2534,7 +2730,7 @@ function loadConfig() {
   return resolveTemplatesInConfig(inherited, configPath);
 }
 function getCustomCommandNames(config) {
-  const reserved = /* @__PURE__ */ new Set(["global", "complete", "agent", "prompts", "docker"]);
+  const reserved = /* @__PURE__ */ new Set(["global", "complete", "agent", "image", "speech", "prompts", "docker"]);
   return Object.keys(config).filter((key) => !reserved.has(key));
 }
 function resolveTemplatesInConfig(config, configPath) {
@@ -3528,8 +3724,8 @@ Denied: ${result.reason ?? "by user"}`
   builder.withTextOnlyHandler("acknowledge");
   builder.withTextWithGadgetsHandler({
     gadgetName: "TellUser",
-    parameterMapping: (text) => ({ message: text, done: false, type: "info" }),
-    resultMapping: (text) => `\u2139\uFE0F  ${text}`
+    parameterMapping: (text2) => ({ message: text2, done: false, type: "info" }),
+    resultMapping: (text2) => `\u2139\uFE0F  ${text2}`
   });
   builder.withTrailingMessage(
     (ctx) => [
@@ -3538,7 +3734,19 @@ Denied: ${result.reason ?? "by user"}`
       "Maximize efficiency by batching independent operations in a single response."
     ].join(" ")
   );
-  const agent = builder.ask(prompt);
+  let agent;
+  if (options.image || options.audio) {
+    const parts = [text(prompt)];
+    if (options.image) {
+      parts.push(await readImageFile(options.image));
+    }
+    if (options.audio) {
+      parts.push(await readAudioFile(options.audio));
+    }
+    agent = builder.askWithContent(parts);
+  } else {
+    agent = builder.ask(prompt);
+  }
   let textBuffer = "";
   const flushTextBuffer = () => {
     if (textBuffer) {
@@ -3613,6 +3821,7 @@ function registerAgentCommand(program, env, config) {
 }
 
 // src/cli/complete-command.ts
+init_input_content();
 init_messages();
 init_model_shortcuts();
 init_constants();
@@ -3624,7 +3833,18 @@ async function executeComplete(promptArg, options, env) {
   if (options.system) {
     builder.addSystem(options.system);
   }
-  builder.addUser(prompt);
+  if (options.image || options.audio) {
+    const parts = [text(prompt)];
+    if (options.image) {
+      parts.push(await readImageFile(options.image));
+    }
+    if (options.audio) {
+      parts.push(await readAudioFile(options.audio));
+    }
+    builder.addUserMultimodal(parts);
+  } else {
+    builder.addUser(prompt);
+  }
   const messages = builder.build();
   const llmLogsBaseDir = resolveLogDir(options.logLlmRequests, "requests");
   let llmSessionDir;
@@ -4129,19 +4349,118 @@ function registerGadgetCommand(program, env) {
   );
 }
 
+// src/cli/image-command.ts
+import { writeFileSync as writeFileSync2 } from "node:fs";
+var DEFAULT_IMAGE_MODEL = "dall-e-3";
+async function executeImage(promptArg, options, env) {
+  const prompt = await resolvePrompt(promptArg, env);
+  const client = env.createClient();
+  const model = options.model;
+  const n = options.count ? Number.parseInt(options.count, 10) : 1;
+  const stderrTTY = env.stderr.isTTY === true;
+  if (!options.quiet && stderrTTY) {
+    env.stderr.write(`${SUMMARY_PREFIX} Generating image with ${model}...
+`);
+  }
+  const result = await client.image.generate({
+    model,
+    prompt,
+    size: options.size,
+    quality: options.quality,
+    n,
+    responseFormat: options.output ? "b64_json" : "url"
+  });
+  if (options.output) {
+    const imageData = result.images[0];
+    if (imageData.b64Json) {
+      const buffer = Buffer.from(imageData.b64Json, "base64");
+      writeFileSync2(options.output, buffer);
+      if (!options.quiet) {
+        env.stderr.write(`${SUMMARY_PREFIX} Image saved to ${options.output}
+`);
+      }
+    } else if (imageData.url) {
+      env.stdout.write(`${imageData.url}
+`);
+    }
+  } else {
+    for (const image of result.images) {
+      if (image.url) {
+        env.stdout.write(`${image.url}
+`);
+      } else if (image.b64Json) {
+        env.stdout.write(image.b64Json);
+      }
+    }
+  }
+  if (!options.quiet && stderrTTY) {
+    const parts = [
+      `${result.images.length} image(s)`,
+      `size: ${result.usage.size}`,
+      `quality: ${result.usage.quality}`
+    ];
+    if (result.cost !== void 0) {
+      parts.push(`cost: ${formatCost(result.cost)}`);
+    }
+    env.stderr.write(`${SUMMARY_PREFIX} ${parts.join(" | ")}
+`);
+  }
+}
+function registerImageCommand(program, env, config) {
+  program.command(COMMANDS.image).description("Generate images from a text prompt.").argument("[prompt]", "Image generation prompt. If omitted, stdin is used when available.").option(
+    OPTION_FLAGS.model,
+    OPTION_DESCRIPTIONS.model,
+    config?.model ?? DEFAULT_IMAGE_MODEL
+  ).option(OPTION_FLAGS.imageSize, OPTION_DESCRIPTIONS.imageSize, config?.size).option(OPTION_FLAGS.imageQuality, OPTION_DESCRIPTIONS.imageQuality, config?.quality).option(OPTION_FLAGS.imageCount, OPTION_DESCRIPTIONS.imageCount, config?.count?.toString()).option(OPTION_FLAGS.imageOutput, OPTION_DESCRIPTIONS.imageOutput, config?.output).option(OPTION_FLAGS.quiet, OPTION_DESCRIPTIONS.quiet, config?.quiet ?? false).action(
+    (prompt, options) => executeAction(() => executeImage(prompt, options, env), env)
+  );
+}
+
 // src/cli/models-command.ts
 import chalk8 from "chalk";
 init_model_shortcuts();
 async function handleModelsCommand(options, env) {
   const client = env.createClient();
-  const models = client.modelRegistry.listModels(options.provider);
+  const showText = options.all || options.text || !options.image && !options.speech;
+  const showImage = options.all || options.image;
+  const showSpeech = options.all || options.speech;
+  const textModels = showText ? client.modelRegistry.listModels(options.provider) : [];
+  const imageModels = showImage ? client.image.listModels().filter((m) => !options.provider || m.provider === options.provider) : [];
+  const speechModels = showSpeech ? client.speech.listModels().filter((m) => !options.provider || m.provider === options.provider) : [];
   if (options.format === "json") {
-    renderJSON(models, env.stdout);
+    renderJSON(textModels, imageModels, speechModels, env.stdout);
   } else {
-    renderTable(models, options.verbose || false, env.stdout);
+    renderAllTables(textModels, imageModels, speechModels, options.verbose || false, env.stdout);
+  }
+}
+function renderAllTables(textModels, imageModels, speechModels, verbose, stream) {
+  const hasAnyModels = textModels.length > 0 || imageModels.length > 0 || speechModels.length > 0;
+  if (!hasAnyModels) {
+    stream.write(chalk8.yellow("\nNo models found matching the specified criteria.\n\n"));
+    return;
+  }
+  stream.write(chalk8.bold.cyan("\nAvailable Models\n"));
+  stream.write(chalk8.cyan("=".repeat(80)) + "\n\n");
+  if (textModels.length > 0) {
+    renderTextTable(textModels, verbose, stream);
+  }
+  if (imageModels.length > 0) {
+    renderImageTable(imageModels, verbose, stream);
+  }
+  if (speechModels.length > 0) {
+    renderSpeechTable(speechModels, verbose, stream);
+  }
+  if (textModels.length > 0) {
+    stream.write(chalk8.bold.magenta("Model Shortcuts\n"));
+    stream.write(chalk8.dim("\u2500".repeat(80)) + "\n");
+    const shortcuts = Object.entries(MODEL_ALIASES).sort((a, b) => a[0].localeCompare(b[0]));
+    for (const [shortcut, fullName] of shortcuts) {
+      stream.write(chalk8.cyan(`  ${shortcut.padEnd(15)}`) + chalk8.dim(" \u2192 ") + chalk8.white(fullName) + "\n");
+    }
+    stream.write("\n");
   }
 }
-function renderTable(models, verbose, stream) {
+function renderTextTable(models, verbose, stream) {
   const grouped = /* @__PURE__ */ new Map();
   for (const model of models) {
     const provider = model.provider;
@@ -4150,13 +4469,13 @@ function renderTable(models, verbose, stream) {
     }
     grouped.get(provider).push(model);
   }
-  stream.write(chalk8.bold.cyan("\nAvailable Models\n"));
-  stream.write(chalk8.cyan("=".repeat(80)) + "\n\n");
+  stream.write(chalk8.bold.blue("\u{1F4DD} Text/LLM Models\n"));
+  stream.write(chalk8.dim("\u2500".repeat(80)) + "\n\n");
   const providers = Array.from(grouped.keys()).sort();
   for (const provider of providers) {
     const providerModels = grouped.get(provider);
     const providerName = provider.charAt(0).toUpperCase() + provider.slice(1);
-    stream.write(chalk8.bold.yellow(`${providerName} Models
+    stream.write(chalk8.bold.yellow(`${providerName}
 `));
     if (verbose) {
       renderVerboseTable(providerModels, stream);
@@ -4165,13 +4484,6 @@ function renderTable(models, verbose, stream) {
     }
     stream.write("\n");
   }
-  stream.write(chalk8.bold.magenta("Model Shortcuts\n"));
-  stream.write(chalk8.dim("\u2500".repeat(80)) + "\n");
-  const shortcuts = Object.entries(MODEL_ALIASES).sort((a, b) => a[0].localeCompare(b[0]));
-  for (const [shortcut, fullName] of shortcuts) {
-    stream.write(chalk8.cyan(`  ${shortcut.padEnd(15)}`) + chalk8.dim(" \u2192 ") + chalk8.white(fullName) + "\n");
-  }
-  stream.write("\n");
 }
 function renderCompactTable(models, stream) {
   const idWidth = 25;
@@ -4248,9 +4560,171 @@ function renderVerboseTable(models, stream) {
   }
   stream.write("\n");
 }
-function renderJSON(models, stream) {
-  const output = {
-    models: models.map((model) => ({
+function renderImageTable(models, verbose, stream) {
+  stream.write(chalk8.bold.green("\u{1F3A8} Image Generation Models\n"));
+  stream.write(chalk8.dim("\u2500".repeat(80)) + "\n\n");
+  const grouped = /* @__PURE__ */ new Map();
+  for (const model of models) {
+    if (!grouped.has(model.provider)) {
+      grouped.set(model.provider, []);
+    }
+    grouped.get(model.provider).push(model);
+  }
+  for (const [provider, providerModels] of Array.from(grouped.entries()).sort()) {
+    const providerName = provider.charAt(0).toUpperCase() + provider.slice(1);
+    stream.write(chalk8.bold.yellow(`${providerName}
+`));
+    if (verbose) {
+      for (const model of providerModels) {
+        stream.write(chalk8.bold.green(`
+  ${model.modelId}
+`));
+        stream.write(chalk8.dim("  " + "\u2500".repeat(60)) + "\n");
+        stream.write(`  ${chalk8.dim("Name:")}      ${chalk8.white(model.displayName)}
+`);
+        stream.write(`  ${chalk8.dim("Sizes:")}     ${chalk8.yellow(model.supportedSizes.join(", "))}
+`);
+        if (model.supportedQualities) {
+          stream.write(`  ${chalk8.dim("Qualities:")} ${chalk8.yellow(model.supportedQualities.join(", "))}
+`);
+        }
+        stream.write(`  ${chalk8.dim("Max Images:")} ${chalk8.yellow(model.maxImages.toString())}
+`);
+        stream.write(`  ${chalk8.dim("Pricing:")}   ${chalk8.cyan(formatImagePrice(model))}
+`);
+        if (model.features) {
+          const features = [];
+          if (model.features.textRendering) features.push("text-rendering");
+          if (model.features.transparency) features.push("transparency");
+          if (model.features.conversational) features.push("conversational");
+          if (features.length > 0) {
+            stream.write(`  ${chalk8.dim("Features:")}  ${chalk8.blue(features.join(", "))}
+`);
+          }
+        }
+      }
+    } else {
+      const idWidth = 32;
+      const nameWidth = 25;
+      const sizesWidth = 20;
+      const priceWidth = 15;
+      stream.write(chalk8.dim("\u2500".repeat(idWidth + nameWidth + sizesWidth + priceWidth + 6)) + "\n");
+      stream.write(
+        chalk8.bold(
+          "Model ID".padEnd(idWidth) + "  " + "Display Name".padEnd(nameWidth) + "  " + "Sizes".padEnd(sizesWidth) + "  " + "Price".padEnd(priceWidth)
+        ) + "\n"
+      );
+      stream.write(chalk8.dim("\u2500".repeat(idWidth + nameWidth + sizesWidth + priceWidth + 6)) + "\n");
+      for (const model of providerModels) {
+        const sizes = model.supportedSizes.length > 2 ? model.supportedSizes.slice(0, 2).join(", ") + "..." : model.supportedSizes.join(", ");
+        stream.write(
+          chalk8.green(model.modelId.padEnd(idWidth)) + "  " + chalk8.white(model.displayName.substring(0, nameWidth - 1).padEnd(nameWidth)) + "  " + chalk8.yellow(sizes.padEnd(sizesWidth)) + "  " + chalk8.cyan(formatImagePrice(model).padEnd(priceWidth)) + "\n"
+        );
+      }
+      stream.write(chalk8.dim("\u2500".repeat(idWidth + nameWidth + sizesWidth + priceWidth + 6)) + "\n");
+    }
+    stream.write("\n");
+  }
+}
+function renderSpeechTable(models, verbose, stream) {
+  stream.write(chalk8.bold.magenta("\u{1F3A4} Speech (TTS) Models\n"));
+  stream.write(chalk8.dim("\u2500".repeat(80)) + "\n\n");
+  const grouped = /* @__PURE__ */ new Map();
+  for (const model of models) {
+    if (!grouped.has(model.provider)) {
+      grouped.set(model.provider, []);
+    }
+    grouped.get(model.provider).push(model);
+  }
+  for (const [provider, providerModels] of Array.from(grouped.entries()).sort()) {
+    const providerName = provider.charAt(0).toUpperCase() + provider.slice(1);
+    stream.write(chalk8.bold.yellow(`${providerName}
+`));
+    if (verbose) {
+      for (const model of providerModels) {
+        stream.write(chalk8.bold.green(`
+  ${model.modelId}
+`));
+        stream.write(chalk8.dim("  " + "\u2500".repeat(60)) + "\n");
+        stream.write(`  ${chalk8.dim("Name:")}    ${chalk8.white(model.displayName)}
+`);
+        stream.write(`  ${chalk8.dim("Voices:")}  ${chalk8.yellow(model.voices.length.toString())} voices
+`);
+        if (model.voices.length <= 6) {
+          stream.write(`            ${chalk8.dim(model.voices.join(", "))}
+`);
+        } else {
+          stream.write(`            ${chalk8.dim(model.voices.slice(0, 6).join(", ") + "...")}
+`);
+        }
+        stream.write(`  ${chalk8.dim("Formats:")} ${chalk8.yellow(model.formats.join(", "))}
+`);
+        stream.write(`  ${chalk8.dim("Max Input:")} ${chalk8.yellow(model.maxInputLength.toString())} chars
+`);
+        stream.write(`  ${chalk8.dim("Pricing:")} ${chalk8.cyan(formatSpeechPrice(model))}
+`);
+        if (model.features) {
+          const features = [];
+          if (model.features.multiSpeaker) features.push("multi-speaker");
+          if (model.features.voiceInstructions) features.push("voice-instructions");
+          if (model.features.languages) features.push(`${model.features.languages} languages`);
+          if (features.length > 0) {
+            stream.write(`  ${chalk8.dim("Features:")} ${chalk8.blue(features.join(", "))}
+`);
+          }
+        }
+      }
+    } else {
+      const idWidth = 30;
+      const nameWidth = 28;
+      const voicesWidth = 12;
+      const priceWidth = 18;
+      stream.write(chalk8.dim("\u2500".repeat(idWidth + nameWidth + voicesWidth + priceWidth + 6)) + "\n");
+      stream.write(
+        chalk8.bold(
+          "Model ID".padEnd(idWidth) + "  " + "Display Name".padEnd(nameWidth) + "  " + "Voices".padEnd(voicesWidth) + "  " + "Price".padEnd(priceWidth)
+        ) + "\n"
+      );
+      stream.write(chalk8.dim("\u2500".repeat(idWidth + nameWidth + voicesWidth + priceWidth + 6)) + "\n");
+      for (const model of providerModels) {
+        stream.write(
+          chalk8.green(model.modelId.padEnd(idWidth)) + "  " + chalk8.white(model.displayName.substring(0, nameWidth - 1).padEnd(nameWidth)) + "  " + chalk8.yellow(`${model.voices.length} voices`.padEnd(voicesWidth)) + "  " + chalk8.cyan(formatSpeechPrice(model).padEnd(priceWidth)) + "\n"
+        );
+      }
+      stream.write(chalk8.dim("\u2500".repeat(idWidth + nameWidth + voicesWidth + priceWidth + 6)) + "\n");
+    }
+    stream.write("\n");
+  }
+}
+function formatImagePrice(model) {
+  if (model.pricing.perImage !== void 0) {
+    return `$${model.pricing.perImage.toFixed(2)}/img`;
+  }
+  if (model.pricing.bySize) {
+    const prices = Object.values(model.pricing.bySize);
+    const minPrice = Math.min(...prices.flatMap((p) => typeof p === "number" ? [p] : Object.values(p)));
+    const maxPrice = Math.max(...prices.flatMap((p) => typeof p === "number" ? [p] : Object.values(p)));
+    if (minPrice === maxPrice) {
+      return `$${minPrice.toFixed(2)}/img`;
+    }
+    return `$${minPrice.toFixed(2)}-${maxPrice.toFixed(2)}`;
+  }
+  return "varies";
+}
+function formatSpeechPrice(model) {
+  if (model.pricing.perCharacter !== void 0) {
+    const perMillion = model.pricing.perCharacter * 1e6;
+    return `$${perMillion.toFixed(0)}/1M chars`;
+  }
+  if (model.pricing.perMinute !== void 0) {
+    return `~$${model.pricing.perMinute.toFixed(2)}/min`;
+  }
+  return "varies";
+}
+function renderJSON(textModels, imageModels, speechModels, stream) {
+  const output = {};
+  if (textModels.length > 0) {
+    output.textModels = textModels.map((model) => ({
       provider: model.provider,
       modelId: model.modelId,
       displayName: model.displayName,
@@ -4266,9 +4740,33 @@ function renderJSON(models, stream) {
       knowledgeCutoff: model.knowledgeCutoff,
       features: model.features,
       metadata: model.metadata
-    })),
-    shortcuts: MODEL_ALIASES
-  };
+    }));
+    output.shortcuts = MODEL_ALIASES;
+  }
+  if (imageModels.length > 0) {
+    output.imageModels = imageModels.map((model) => ({
+      provider: model.provider,
+      modelId: model.modelId,
+      displayName: model.displayName,
+      supportedSizes: model.supportedSizes,
+      supportedQualities: model.supportedQualities,
+      maxImages: model.maxImages,
+      pricing: model.pricing,
+      features: model.features
+    }));
+  }
+  if (speechModels.length > 0) {
+    output.speechModels = speechModels.map((model) => ({
+      provider: model.provider,
+      modelId: model.modelId,
+      displayName: model.displayName,
+      voices: model.voices,
+      formats: model.formats,
+      maxInputLength: model.maxInputLength,
+      pricing: model.pricing,
+      features: model.features
+    }));
+  }
   stream.write(JSON.stringify(output, null, 2) + "\n");
 }
 function formatTokens2(count) {
@@ -4281,7 +4779,7 @@ function formatTokens2(count) {
   }
 }
 function registerModelsCommand(program, env) {
-  program.command(COMMANDS.models).description("List all available LLM models with pricing and capabilities.").option("--provider <name>", "Filter by provider (openai, anthropic, gemini)").option("--format <format>", "Output format: table or json", "table").option("--verbose", "Show detailed model information", false).action(
+  program.command(COMMANDS.models).description("List available models with pricing and capabilities.").option("--provider <name>", "Filter by provider (openai, anthropic, gemini)").option("--format <format>", "Output format: table or json", "table").option("--verbose", "Show detailed model information", false).option("--text", "Show text/LLM models (default if no type specified)").option("--image", "Show image generation models").option("--speech", "Show speech/TTS models").option("--all", "Show all model types (text, image, speech)").action(
     (options) => executeAction(
       () => handleModelsCommand(options, env),
       env
@@ -4289,6 +4787,96 @@ function registerModelsCommand(program, env) {
   );
 }
 
+// src/cli/speech-command.ts
+import { writeFileSync as writeFileSync3 } from "node:fs";
+var DEFAULT_SPEECH_MODEL = "tts-1";
+var DEFAULT_VOICE = "nova";
+async function executeSpeech(textArg, options, env) {
+  const text2 = await resolvePrompt(textArg, env);
+  const client = env.createClient();
+  const model = options.model;
+  const voice = options.voice ?? DEFAULT_VOICE;
+  const speed = options.speed ? Number.parseFloat(options.speed) : void 0;
+  const stderrTTY = env.stderr.isTTY === true;
+  if (!options.quiet && stderrTTY) {
+    env.stderr.write(`${SUMMARY_PREFIX} Generating speech with ${model} (voice: ${voice})...
+`);
+  }
+  const result = await client.speech.generate({
+    model,
+    input: text2,
+    voice,
+    responseFormat: options.format,
+    speed
+  });
+  const audioBuffer = Buffer.from(result.audio);
+  if (options.output) {
+    writeFileSync3(options.output, audioBuffer);
+    if (!options.quiet) {
+      env.stderr.write(`${SUMMARY_PREFIX} Audio saved to ${options.output}
+`);
+    }
+  } else {
+    env.stdout.write(audioBuffer);
+  }
+  if (!options.quiet && stderrTTY) {
+    const parts = [
+      `${result.usage.characterCount} characters`,
+      `format: ${result.format}`
+    ];
+    if (result.cost !== void 0) {
+      parts.push(`cost: ${formatCost(result.cost)}`);
+    }
+    env.stderr.write(`${SUMMARY_PREFIX} ${parts.join(" | ")}
+`);
+  }
+}
+function registerSpeechCommand(program, env, config) {
+  program.command(COMMANDS.speech).description("Generate speech audio from text.").argument("[text]", "Text to convert to speech. If omitted, stdin is used when available.").option(
+    OPTION_FLAGS.model,
+    OPTION_DESCRIPTIONS.model,
+    config?.model ?? DEFAULT_SPEECH_MODEL
+  ).option(OPTION_FLAGS.voice, OPTION_DESCRIPTIONS.voice, config?.voice ?? DEFAULT_VOICE).option(OPTION_FLAGS.speechFormat, OPTION_DESCRIPTIONS.speechFormat, config?.format).option(OPTION_FLAGS.speechSpeed, OPTION_DESCRIPTIONS.speechSpeed, config?.speed?.toString()).option(OPTION_FLAGS.speechOutput, OPTION_DESCRIPTIONS.speechOutput, config?.output).option(OPTION_FLAGS.quiet, OPTION_DESCRIPTIONS.quiet, config?.quiet ?? false).action(
+    (text2, options) => executeAction(() => executeSpeech(text2, options, env), env)
+  );
+}
+
+// src/cli/vision-command.ts
+init_model_shortcuts();
+async function executeVision(imagePath, options, env) {
+  const client = env.createClient();
+  const model = resolveModel(options.model);
+  const imageBuffer = await readFileBuffer(imagePath);
+  const prompt = options.prompt ?? "Describe this image in detail.";
+  const stderrTTY = env.stderr.isTTY === true;
+  if (!options.quiet && stderrTTY) {
+    env.stderr.write(`${SUMMARY_PREFIX} Analyzing image with ${model}...
+`);
+  }
+  const result = await client.vision.analyze({
+    model,
+    image: imageBuffer,
+    prompt,
+    maxTokens: options.maxTokens
+  });
+  env.stdout.write(result);
+  env.stdout.write("\n");
+}
+function registerVisionCommand(program, env) {
+  program.command(COMMANDS.vision ?? "vision").description("Analyze an image using vision-capable models").argument("<image>", "Path to image file to analyze").option(
+    OPTION_FLAGS.model,
+    OPTION_DESCRIPTIONS.model,
+    "gpt-4o"
+    // Default to a vision-capable model
+  ).option("-p, --prompt <prompt>", "Analysis prompt describing what to extract or describe").option(
+    OPTION_FLAGS.maxTokens,
+    OPTION_DESCRIPTIONS.maxTokens,
+    createNumericParser({ label: "Max tokens", integer: true, min: 1 })
+  ).option(OPTION_FLAGS.quiet, OPTION_DESCRIPTIONS.quiet).action(
+    (imagePath, options) => executeAction(() => executeVision(imagePath, options, env), env)
+  );
+}
+
 // src/cli/environment.ts
 init_client();
 init_logger();
@@ -4334,7 +4922,7 @@ function createLoggerFactory(config) {
 }
 function createPromptFunction(stdin, stdout) {
   return (question) => {
-    return new Promise((resolve2) => {
+    return new Promise((resolve3) => {
       const rl = readline.createInterface({
         input: stdin,
         output: stdout
@@ -4349,7 +4937,7 @@ function createPromptFunction(stdin, stdout) {
 `);
       rl.question(chalk9.green.bold("You: "), (answer) => {
         rl.close();
-        resolve2(answer);
+        resolve3(answer);
       });
     });
   };
@@ -4440,6 +5028,9 @@ function createProgram(env, config) {
   });
   registerCompleteCommand(program, env, config?.complete);
   registerAgentCommand(program, env, config?.agent);
+  registerImageCommand(program, env, config?.image);
+  registerSpeechCommand(program, env, config?.speech);
+  registerVisionCommand(program, env);
   registerModelsCommand(program, env);
   registerGadgetCommand(program, env);
   if (config) {