llmist 2.4.0 → 2.5.0

package/dist/cli.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import "./chunk-QFRVTS5F.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-6ZDUWO6N.js";
+} from "./chunk-YHS2DYXP.js";
 
 // src/cli/constants.ts
 var CLI_NAME = "llmist";
@@ -38,7 +45,8 @@ var COMMANDS = {
   models: "models",
   gadget: "gadget",
   image: "image",
-  speech: "speech"
+  speech: "speech",
+  vision: "vision"
 };
 var LOG_LEVELS = ["silly", "trace", "debug", "info", "warn", "error", "fatal"];
 var DEFAULT_MODEL = "openai:gpt-5-nano";
@@ -60,6 +68,9 @@ var OPTION_FLAGS = {
   dockerRo: "--docker-ro",
   noDocker: "--no-docker",
   dockerDev: "--docker-dev",
+  // Multimodal input options
+  inputImage: "--image <path>",
+  inputAudio: "--audio <path>",
   // Image generation options
   imageSize: "--size <size>",
   imageQuality: "--quality <quality>",
@@ -85,6 +96,9 @@ 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).",
@@ -108,7 +122,7 @@ import { Command, InvalidArgumentError as InvalidArgumentError2 } from "commande
 // package.json
 var package_default = {
   name: "llmist",
-  version: "2.4.0",
+  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",
@@ -248,6 +262,7 @@ function isAbortError(error) {
 }
 
 // src/cli/agent-command.ts
+init_input_content();
 init_registry();
 init_constants();
 
@@ -572,6 +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";
@@ -735,7 +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({
@@ -1006,7 +1090,7 @@ error: ${message}`;
 // src/cli/builtins/index.ts
 var builtinGadgetRegistry = {
   ListDirectory: listDirectory,
-  ReadFile: readFile,
+  ReadFile: readFile2,
   WriteFile: writeFile,
   EditFile: editFile,
   RunCommand: runCommand
@@ -1141,6 +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";
@@ -1158,7 +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");
@@ -1232,9 +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();
 }
@@ -1248,8 +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}
@@ -1417,12 +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.
@@ -1901,7 +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 = []) => [
@@ -1925,7 +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 = {};
@@ -3639,8 +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) => [
@@ -3649,7 +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) {
@@ -3724,6 +3821,7 @@ function registerAgentCommand(program, env, config) {
 }
 
 // src/cli/complete-command.ts
+init_input_content();
 init_messages();
 init_model_shortcuts();
 init_constants();
@@ -3735,7 +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;
@@ -4683,7 +4792,7 @@ import { writeFileSync as writeFileSync3 } from "node:fs";
 var DEFAULT_SPEECH_MODEL = "tts-1";
 var DEFAULT_VOICE = "nova";
 async function executeSpeech(textArg, options, env) {
-  const text = await resolvePrompt(textArg, env);
+  const text2 = await resolvePrompt(textArg, env);
   const client = env.createClient();
   const model = options.model;
   const voice = options.voice ?? DEFAULT_VOICE;
@@ -4695,7 +4804,7 @@ async function executeSpeech(textArg, options, env) {
   }
   const result = await client.speech.generate({
     model,
-    input: text,
+    input: text2,
     voice,
     responseFormat: options.format,
     speed
@@ -4728,7 +4837,43 @@ function registerSpeechCommand(program, env, config) {
     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(
-    (text, options) => executeAction(() => executeSpeech(text, options, env), env)
+    (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)
   );
 }
 
@@ -4777,7 +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
@@ -4792,7 +4937,7 @@ function createPromptFunction(stdin, stdout) {
 `);
       rl.question(chalk9.green.bold("You: "), (answer) => {
         rl.close();
-        resolve2(answer);
+        resolve3(answer);
       });
     });
   };
@@ -4885,6 +5030,7 @@ function createProgram(env, config) {
   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) {