@steipete/summarize 0.3.0 → 0.4.0

package/dist/cli.cjs

@@ -1196,8 +1196,8 @@ var require_command = __commonJS({
   "node_modules/.pnpm/commander@14.0.2/node_modules/commander/lib/command.js"(exports2) {
     var EventEmitter = require("node:events").EventEmitter;
     var childProcess = require("node:child_process");
-    var path7 = require("node:path");
-    var fs6 = require("node:fs");
+    var path8 = require("node:path");
+    var fs7 = require("node:fs");
     var process14 = require("node:process");
     var { Argument: Argument2, humanReadableArgName } = require_argument();
     var { CommanderError: CommanderError2 } = require_error();
@@ -2191,7 +2191,7 @@ Expecting one of '${allowedValues.join("', '")}'`);
        * @param {string} subcommandName
        */
       _checkForMissingExecutable(executableFile, executableDir, subcommandName) {
-        if (fs6.existsSync(executableFile)) return;
+        if (fs7.existsSync(executableFile)) return;
         const executableDirMessage = executableDir ? `searched for local subcommand relative to directory '${executableDir}'` : "no directory for search for local subcommand, use .executableDir() to supply a custom directory";
         const executableMissing = `'${executableFile}' does not exist
  - if '${subcommandName}' is not meant to be an executable command, remove description parameter from '.command()' and use '.description()' instead
@@ -2209,11 +2209,11 @@ Expecting one of '${allowedValues.join("', '")}'`);
         let launchWithNode = false;
         const sourceExt = [".js", ".ts", ".tsx", ".mjs", ".cjs"];
         function findFile(baseDir, baseName) {
-          const localBin = path7.resolve(baseDir, baseName);
-          if (fs6.existsSync(localBin)) return localBin;
-          if (sourceExt.includes(path7.extname(baseName))) return void 0;
+          const localBin = path8.resolve(baseDir, baseName);
+          if (fs7.existsSync(localBin)) return localBin;
+          if (sourceExt.includes(path8.extname(baseName))) return void 0;
           const foundExt = sourceExt.find(
-            (ext) => fs6.existsSync(`${localBin}${ext}`)
+            (ext) => fs7.existsSync(`${localBin}${ext}`)
           );
           if (foundExt) return `${localBin}${foundExt}`;
           return void 0;
@@ -2225,21 +2225,21 @@ Expecting one of '${allowedValues.join("', '")}'`);
         if (this._scriptPath) {
           let resolvedScriptPath;
           try {
-            resolvedScriptPath = fs6.realpathSync(this._scriptPath);
+            resolvedScriptPath = fs7.realpathSync(this._scriptPath);
           } catch {
             resolvedScriptPath = this._scriptPath;
           }
-          executableDir = path7.resolve(
-            path7.dirname(resolvedScriptPath),
+          executableDir = path8.resolve(
+            path8.dirname(resolvedScriptPath),
             executableDir
           );
         }
         if (executableDir) {
           let localFile = findFile(executableDir, executableFile);
           if (!localFile && !subcommand._executableFile && this._scriptPath) {
-            const legacyName = path7.basename(
+            const legacyName = path8.basename(
               this._scriptPath,
-              path7.extname(this._scriptPath)
+              path8.extname(this._scriptPath)
             );
             if (legacyName !== this._name) {
               localFile = findFile(
@@ -2250,7 +2250,7 @@ Expecting one of '${allowedValues.join("', '")}'`);
           }
           executableFile = localFile || executableFile;
         }
-        launchWithNode = sourceExt.includes(path7.extname(executableFile));
+        launchWithNode = sourceExt.includes(path8.extname(executableFile));
         let proc;
         if (process14.platform !== "win32") {
           if (launchWithNode) {
@@ -3165,7 +3165,7 @@ Expecting one of '${allowedValues.join("', '")}'`);
        * @return {Command}
        */
       nameFromFilename(filename) {
-        this._name = path7.basename(filename, path7.extname(filename));
+        this._name = path8.basename(filename, path8.extname(filename));
         return this;
       }
       /**
@@ -3179,9 +3179,9 @@ Expecting one of '${allowedValues.join("', '")}'`);
        * @param {string} [path]
        * @return {(string|null|Command)}
        */
-      executableDir(path8) {
-        if (path8 === void 0) return this._executableDir;
-        this._executableDir = path8;
+      executableDir(path9) {
+        if (path9 === void 0) return this._executableDir;
+        this._executableDir = path9;
         return this;
       }
       /**
@@ -4902,17 +4902,17 @@ var require_request = __commonJS({
     function buildUrl(id, options = {}) {
       var _a18, _b16;
       const method = ((_a18 = options.method) !== null && _a18 !== void 0 ? _a18 : "post").toLowerCase();
-      const path7 = ((_b16 = options.path) !== null && _b16 !== void 0 ? _b16 : "").replace(/^\//, "").replace(/\/{2,}/, "/");
+      const path8 = ((_b16 = options.path) !== null && _b16 !== void 0 ? _b16 : "").replace(/^\//, "").replace(/\/{2,}/, "/");
       const input = options.input;
       const params = Object.assign(Object.assign({}, options.query || {}), method === "get" ? input : {});
       const queryParams = Object.keys(params).length > 0 ? `?${new URLSearchParams(params).toString()}` : "";
       if ((0, utils_1.isValidUrl)(id)) {
         const url3 = id.endsWith("/") ? id : `${id}/`;
-        return `${url3}${path7}${queryParams}`;
+        return `${url3}${path8}${queryParams}`;
       }
       const appId = (0, utils_1.ensureEndpointIdFormat)(id);
       const subdomain = options.subdomain ? `${options.subdomain}.` : "";
-      const url2 = `https://${subdomain}fal.run/${appId}/${path7}`;
+      const url2 = `https://${subdomain}fal.run/${appId}/${path8}`;
       return `${url2.replace(/\/$/, "")}${queryParams}`;
     }
   }
@@ -8709,10 +8709,10 @@ function mergeDefs(...defs) {
 function cloneDef(schema) {
   return mergeDefs(schema._zod.def);
 }
-function getElementAtPath(obj, path7) {
-  if (!path7)
+function getElementAtPath(obj, path8) {
+  if (!path8)
     return obj;
-  return path7.reduce((acc, key) => acc?.[key], obj);
+  return path8.reduce((acc, key) => acc?.[key], obj);
 }
 function promiseAllObject(promisesObj) {
   const keys = Object.keys(promisesObj);
@@ -9007,11 +9007,11 @@ function aborted(x, startIndex = 0) {
   }
   return false;
 }
-function prefixIssues(path7, issues) {
+function prefixIssues(path8, issues) {
   return issues.map((iss) => {
     var _a18;
     (_a18 = iss).path ?? (_a18.path = []);
-    iss.path.unshift(path7);
+    iss.path.unshift(path8);
     return iss;
   });
 }
@@ -9232,7 +9232,7 @@ function formatError(error47, mapper = (issue2) => issue2.message) {
 }
 function treeifyError(error47, mapper = (issue2) => issue2.message) {
   const result = { errors: [] };
-  const processError = (error48, path7 = []) => {
+  const processError = (error48, path8 = []) => {
     var _a18, _b16;
     for (const issue2 of error48.issues) {
       if (issue2.code === "invalid_union" && issue2.errors.length) {
@@ -9242,7 +9242,7 @@ function treeifyError(error47, mapper = (issue2) => issue2.message) {
       } else if (issue2.code === "invalid_element") {
         processError({ issues: issue2.issues }, issue2.path);
       } else {
-        const fullpath = [...path7, ...issue2.path];
+        const fullpath = [...path8, ...issue2.path];
         if (fullpath.length === 0) {
           result.errors.push(mapper(issue2));
           continue;
@@ -9274,8 +9274,8 @@ function treeifyError(error47, mapper = (issue2) => issue2.message) {
 }
 function toDotPath(_path) {
   const segs = [];
-  const path7 = _path.map((seg) => typeof seg === "object" ? seg.key : seg);
-  for (const seg of path7) {
+  const path8 = _path.map((seg) => typeof seg === "object" ? seg.key : seg);
+  for (const seg of path8) {
     if (typeof seg === "number")
       segs.push(`[${seg}]`);
     else if (typeof seg === "symbol")
@@ -21848,13 +21848,13 @@ function resolveRef(ref, ctx) {
   if (!ref.startsWith("#")) {
     throw new Error("External $ref is not supported, only local refs (#/...) are allowed");
   }
-  const path7 = ref.slice(1).split("/").filter(Boolean);
-  if (path7.length === 0) {
+  const path8 = ref.slice(1).split("/").filter(Boolean);
+  if (path8.length === 0) {
     return ctx.rootSchema;
   }
   const defsKey = ctx.version === "draft-2020-12" ? "$defs" : "definitions";
-  if (path7[0] === defsKey) {
-    const key = path7[1];
+  if (path8[0] === defsKey) {
+    const key = path8[1];
     if (!key || !ctx.defs[key]) {
       throw new Error(`Reference not found: ${ref}`);
     }
@@ -22930,8 +22930,8 @@ var init_parseUtil = __esm({
     init_errors3();
     init_en2();
     makeIssue = (params) => {
-      const { data, path: path7, errorMaps, issueData } = params;
-      const fullPath = [...path7, ...issueData.path || []];
+      const { data, path: path8, errorMaps, issueData } = params;
+      const fullPath = [...path8, ...issueData.path || []];
       const fullIssue = {
         ...issueData,
         path: fullPath
@@ -23211,11 +23211,11 @@ var init_types = __esm({
     init_parseUtil();
     init_util2();
     ParseInputLazyPath = class {
-      constructor(parent, value, path7, key) {
+      constructor(parent, value, path8, key) {
         this._cachedPath = [];
         this.parent = parent;
         this.data = value;
-        this._path = path7;
+        this._path = path8;
         this._key = key;
       }
       get path() {
@@ -28921,8 +28921,8 @@ var require_token_util = __commonJS({
       saveToken: () => saveToken
     });
     module2.exports = __toCommonJS(token_util_exports);
-    var path7 = __toESM2(require("path"));
-    var fs6 = __toESM2(require("fs"));
+    var path8 = __toESM2(require("path"));
+    var fs7 = __toESM2(require("fs"));
     var import_token_error = require_token_error();
     var import_token_io = require_token_io();
     function getVercelDataDir() {
@@ -28931,18 +28931,18 @@ var require_token_util = __commonJS({
       if (!dataDir) {
         return null;
       }
-      return path7.join(dataDir, vercelFolder);
+      return path8.join(dataDir, vercelFolder);
     }
     function getVercelCliToken() {
       const dataDir = getVercelDataDir();
       if (!dataDir) {
         return null;
       }
-      const tokenPath = path7.join(dataDir, "auth.json");
-      if (!fs6.existsSync(tokenPath)) {
+      const tokenPath = path8.join(dataDir, "auth.json");
+      if (!fs7.existsSync(tokenPath)) {
         return null;
       }
-      const token = fs6.readFileSync(tokenPath, "utf8");
+      const token = fs7.readFileSync(tokenPath, "utf8");
       if (!token) {
         return null;
       }
@@ -28983,11 +28983,11 @@ var require_token_util = __commonJS({
         throw new import_token_error.VercelOidcTokenError("Unable to find root directory");
       }
       try {
-        const prjPath = path7.join(dir, ".vercel", "project.json");
-        if (!fs6.existsSync(prjPath)) {
+        const prjPath = path8.join(dir, ".vercel", "project.json");
+        if (!fs7.existsSync(prjPath)) {
           throw new import_token_error.VercelOidcTokenError("project.json not found");
         }
-        const prj = JSON.parse(fs6.readFileSync(prjPath, "utf8"));
+        const prj = JSON.parse(fs7.readFileSync(prjPath, "utf8"));
         if (typeof prj.projectId !== "string" && typeof prj.orgId !== "string") {
           throw new TypeError("Expected a string-valued projectId property");
         }
@@ -29002,11 +29002,11 @@ var require_token_util = __commonJS({
         if (!dir) {
           throw new import_token_error.VercelOidcTokenError("Unable to find user data directory");
         }
-        const tokenPath = path7.join(dir, "com.vercel.token", `${projectId}.json`);
+        const tokenPath = path8.join(dir, "com.vercel.token", `${projectId}.json`);
         const tokenJson = JSON.stringify(token);
-        fs6.mkdirSync(path7.dirname(tokenPath), { mode: 504, recursive: true });
-        fs6.writeFileSync(tokenPath, tokenJson);
-        fs6.chmodSync(tokenPath, 432);
+        fs7.mkdirSync(path8.dirname(tokenPath), { mode: 504, recursive: true });
+        fs7.writeFileSync(tokenPath, tokenJson);
+        fs7.chmodSync(tokenPath, 432);
         return;
       } catch (e) {
         throw new import_token_error.VercelOidcTokenError(`Failed to save token`, e);
@@ -29018,11 +29018,11 @@ var require_token_util = __commonJS({
         if (!dir) {
           return null;
         }
-        const tokenPath = path7.join(dir, "com.vercel.token", `${projectId}.json`);
-        if (!fs6.existsSync(tokenPath)) {
+        const tokenPath = path8.join(dir, "com.vercel.token", `${projectId}.json`);
+        if (!fs7.existsSync(tokenPath)) {
           return null;
         }
-        const token = JSON.parse(fs6.readFileSync(tokenPath, "utf8"));
+        const token = JSON.parse(fs7.readFileSync(tokenPath, "utf8"));
         assertVercelOidcTokenResponse(token);
         return token;
       } catch (e) {
@@ -42982,7 +42982,7 @@ function createXai(options = {}) {
   const createImageModel = (modelId) => {
     return new OpenAICompatibleImageModel(modelId, {
       provider: "xai.image",
-      url: ({ path: path7 }) => `${baseURL}${path7}`,
+      url: ({ path: path8 }) => `${baseURL}${path8}`,
       headers: getHeaders,
       fetch: options.fetch,
       errorStructure: xaiErrorStructure
@@ -51227,37 +51227,37 @@ function createOpenAI(options = {}) {
   );
   const createChatModel = (modelId) => new OpenAIChatLanguageModel(modelId, {
     provider: `${providerName}.chat`,
-    url: ({ path: path7 }) => `${baseURL}${path7}`,
+    url: ({ path: path8 }) => `${baseURL}${path8}`,
     headers: getHeaders,
     fetch: options.fetch
   });
   const createCompletionModel = (modelId) => new OpenAICompletionLanguageModel(modelId, {
     provider: `${providerName}.completion`,
-    url: ({ path: path7 }) => `${baseURL}${path7}`,
+    url: ({ path: path8 }) => `${baseURL}${path8}`,
     headers: getHeaders,
     fetch: options.fetch
   });
   const createEmbeddingModel = (modelId) => new OpenAIEmbeddingModel(modelId, {
     provider: `${providerName}.embedding`,
-    url: ({ path: path7 }) => `${baseURL}${path7}`,
+    url: ({ path: path8 }) => `${baseURL}${path8}`,
     headers: getHeaders,
     fetch: options.fetch
   });
   const createImageModel = (modelId) => new OpenAIImageModel(modelId, {
     provider: `${providerName}.image`,
-    url: ({ path: path7 }) => `${baseURL}${path7}`,
+    url: ({ path: path8 }) => `${baseURL}${path8}`,
     headers: getHeaders,
     fetch: options.fetch
   });
   const createTranscriptionModel = (modelId) => new OpenAITranscriptionModel(modelId, {
     provider: `${providerName}.transcription`,
-    url: ({ path: path7 }) => `${baseURL}${path7}`,
+    url: ({ path: path8 }) => `${baseURL}${path8}`,
     headers: getHeaders,
     fetch: options.fetch
   });
   const createSpeechModel = (modelId) => new OpenAISpeechModel(modelId, {
     provider: `${providerName}.speech`,
-    url: ({ path: path7 }) => `${baseURL}${path7}`,
+    url: ({ path: path8 }) => `${baseURL}${path8}`,
     headers: getHeaders,
     fetch: options.fetch
   });
@@ -51272,7 +51272,7 @@ function createOpenAI(options = {}) {
   const createResponsesModel = (modelId) => {
     return new OpenAIResponsesLanguageModel(modelId, {
       provider: `${providerName}.responses`,
-      url: ({ path: path7 }) => `${baseURL}${path7}`,
+      url: ({ path: path8 }) => `${baseURL}${path8}`,
       headers: getHeaders,
       fetch: options.fetch,
       fileIdPrefixes: ["file-"]
@@ -55600,8 +55600,8 @@ var init_dist11 = __esm({
 // src/run.ts
 var import_node_child_process2 = require("node:child_process");
 var import_node_fs6 = require("node:fs");
-var import_promises4 = __toESM(require("node:fs/promises"), 1);
-var import_node_path7 = __toESM(require("node:path"), 1);
+var import_promises5 = __toESM(require("node:fs/promises"), 1);
+var import_node_path8 = __toESM(require("node:path"), 1);
 
 // node_modules/.pnpm/commander@14.0.2/node_modules/commander/esm.mjs
 var import_index = __toESM(require_commander(), 1);
@@ -64632,7 +64632,7 @@ function transformGfmAutolinkLiterals(tree) {
     { ignore: ["link", "linkReference"] }
   );
 }
-function findUrl(_, protocol, domain3, path7, match) {
+function findUrl(_, protocol, domain3, path8, match) {
   let prefix = "";
   if (!previous2(match)) {
     return false;
@@ -64645,7 +64645,7 @@ function findUrl(_, protocol, domain3, path7, match) {
   if (!isCorrectDomain(domain3)) {
     return false;
   }
-  const parts = splitUrl(domain3 + path7);
+  const parts = splitUrl(domain3 + path8);
   if (!parts[0]) return false;
   const result = {
     type: "link",
@@ -67598,25 +67598,25 @@ function isRecord2(value) {
 function loadSummarizeConfig({ env: env3 }) {
   const home = env3.HOME?.trim() || (0, import_node_os3.homedir)();
   if (!home) return { config: null, path: null };
-  const path7 = (0, import_node_path.join)(home, ".summarize", "config.json");
+  const path8 = (0, import_node_path.join)(home, ".summarize", "config.json");
   let raw;
   try {
-    raw = (0, import_node_fs.readFileSync)(path7, "utf8");
+    raw = (0, import_node_fs.readFileSync)(path8, "utf8");
   } catch {
-    return { config: null, path: path7 };
+    return { config: null, path: path8 };
   }
   let parsed;
   try {
     parsed = JSON.parse(raw);
   } catch (error47) {
     const message = error47 instanceof Error ? error47.message : String(error47);
-    throw new Error(`Invalid JSON in config file ${path7}: ${message}`);
+    throw new Error(`Invalid JSON in config file ${path8}: ${message}`);
   }
   if (!isRecord2(parsed)) {
-    throw new Error(`Invalid config file ${path7}: expected an object at the top level`);
+    throw new Error(`Invalid config file ${path8}: expected an object at the top level`);
   }
   const model = typeof parsed.model === "string" ? parsed.model : void 0;
-  return { config: { model }, path: path7 };
+  return { config: { model }, path: path8 };
 }
 
 // src/content/asset.ts
@@ -71757,12 +71757,12 @@ var Mime = class {
     }
     return this;
   }
-  getType(path7) {
-    if (typeof path7 !== "string")
+  getType(path8) {
+    if (typeof path8 !== "string")
       return null;
-    const last = path7.replace(/^.*[/\\]/s, "").toLowerCase();
+    const last = path8.replace(/^.*[/\\]/s, "").toLowerCase();
     const ext = last.replace(/^.*\./s, "").toLowerCase();
-    const hasPath = last.length < path7.length;
+    const hasPath = last.length < path8.length;
     const hasDot = ext.length < last.length - 1;
     if (!hasDot && hasPath)
       return null;
@@ -72458,9 +72458,9 @@ var fetchTranscriptFromTranscriptEndpoint = async (fetchImpl, {
     return null;
   }
 };
-function getNestedProperty(object3, path7) {
+function getNestedProperty(object3, path8) {
   let current = object3;
-  for (const key of path7) {
+  for (const key of path8) {
     if (!(isRecord3(current) && key in current)) {
       return null;
     }
@@ -72468,8 +72468,8 @@ function getNestedProperty(object3, path7) {
   }
   return current;
 }
-function getArrayProperty(object3, path7) {
-  const value = getNestedProperty(object3, path7);
+function getArrayProperty(object3, path8) {
+  const value = getNestedProperty(object3, path8);
   return Array.isArray(value) ? value : null;
 }
 var extractTranscriptFromTranscriptEndpoint = (data) => {
@@ -74662,7 +74662,21 @@ function parseFirecrawlMode(raw) {
 function parseMarkdownMode(raw) {
   const normalized = raw.trim().toLowerCase();
   if (normalized === "off" || normalized === "auto" || normalized === "llm") return normalized;
-  throw new Error(`Unsupported --markdown: ${raw}`);
+  throw new Error(`Unsupported --markdown-mode: ${raw}`);
+}
+function parseExtractFormat(raw) {
+  const normalized = raw.trim().toLowerCase();
+  if (normalized === "text" || normalized === "txt" || normalized === "plain") return "text";
+  if (normalized === "md" || normalized === "markdown") return "markdown";
+  throw new Error(`Unsupported --format: ${raw}`);
+}
+function parsePreprocessMode(raw) {
+  const normalized = raw.trim().toLowerCase();
+  if (normalized === "off" || normalized === "auto" || normalized === "always") {
+    return normalized;
+  }
+  if (normalized === "on") return "always";
+  throw new Error(`Unsupported --preprocess: ${raw}`);
 }
 function parseStreamMode(raw) {
   const normalized = raw.trim().toLowerCase();
@@ -75336,37 +75350,102 @@ function createHtmlToMarkdownConverter({
   };
 }
 
+// src/markitdown.ts
+var import_promises3 = __toESM(require("node:fs/promises"), 1);
+var import_node_os5 = require("node:os");
+var import_node_path4 = __toESM(require("node:path"), 1);
+function guessExtension({
+  filenameHint,
+  mediaType
+}) {
+  const ext = filenameHint ? import_node_path4.default.extname(filenameHint).toLowerCase() : "";
+  if (ext) return ext;
+  if (mediaType === "text/html" || mediaType === "application/xhtml+xml") return ".html";
+  if (mediaType === "application/pdf") return ".pdf";
+  return ".bin";
+}
+async function execFileText(execFileImpl, cmd, args, options) {
+  return await new Promise((resolve2, reject) => {
+    execFileImpl(cmd, args, options, (error47, stdout, stderr) => {
+      if (error47) {
+        const stderrText2 = typeof stderr === "string" ? stderr : stderr.toString("utf8");
+        const message = stderrText2.trim() ? `${error47.message}: ${stderrText2.trim()}` : error47.message;
+        reject(new Error(message, { cause: error47 }));
+        return;
+      }
+      const stdoutText = typeof stdout === "string" ? stdout : stdout.toString("utf8");
+      const stderrText = typeof stderr === "string" ? stderr : stderr.toString("utf8");
+      resolve2({ stdout: stdoutText, stderr: stderrText });
+    });
+  });
+}
+async function convertToMarkdownWithMarkitdown({
+  bytes,
+  filenameHint,
+  mediaTypeHint,
+  uvxCommand,
+  timeoutMs,
+  env: env3,
+  execFileImpl
+}) {
+  const dir = await import_promises3.default.mkdtemp(import_node_path4.default.join((0, import_node_os5.tmpdir)(), "summarize-markitdown-"));
+  const ext = guessExtension({ filenameHint, mediaType: mediaTypeHint });
+  const base2 = (filenameHint ? import_node_path4.default.basename(filenameHint, import_node_path4.default.extname(filenameHint)) : "input").replaceAll(/[^\w.-]+/g, "-").slice(0, 64);
+  const filePath = import_node_path4.default.join(dir, `${base2}${ext}`);
+  try {
+    await import_promises3.default.writeFile(filePath, bytes);
+    const from = "markitdown[all]";
+    const { stdout } = await execFileText(
+      execFileImpl,
+      uvxCommand && uvxCommand.trim().length > 0 ? uvxCommand.trim() : "uvx",
+      ["--from", from, "markitdown", filePath],
+      {
+        timeout: timeoutMs,
+        env: { ...process.env, ...env3 },
+        maxBuffer: 50 * 1024 * 1024
+      }
+    );
+    const markdown = stdout.trim();
+    if (!markdown) {
+      throw new Error("markitdown returned empty output");
+    }
+    return markdown;
+  } finally {
+    await import_promises3.default.rm(dir, { recursive: true, force: true });
+  }
+}
+
 // src/pricing/litellm.ts
-var import_node_path5 = __toESM(require("node:path"), 1);
+var import_node_path6 = __toESM(require("node:path"), 1);
 
 // node_modules/.pnpm/tokentally@https+++codeload.github.com+steipete+tokentally+tar.gz+99865e5c16f5340c9589a2c5d85c3ea47dbcec82/node_modules/tokentally/dist/node/litellm.js
 var import_node_fs4 = require("node:fs");
-var import_promises3 = __toESM(require("node:fs/promises"), 1);
-var import_node_path4 = __toESM(require("node:path"), 1);
+var import_promises4 = __toESM(require("node:fs/promises"), 1);
+var import_node_path5 = __toESM(require("node:path"), 1);
 var LITELLM_CATALOG_URL = "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json";
 var CACHE_TTL_MS = 7 * 24 * 60 * 60 * 1e3;
 function cachePaths(env3) {
   const override = env3.TOKENTALLY_CACHE_DIR?.trim();
   const home = env3.HOME?.trim();
-  const cacheDir = override && override.length > 0 ? override : home ? import_node_path4.default.join(home, ".tokentally", "cache") : null;
+  const cacheDir = override && override.length > 0 ? override : home ? import_node_path5.default.join(home, ".tokentally", "cache") : null;
   if (!cacheDir)
     return null;
   return {
-    catalogPath: import_node_path4.default.join(cacheDir, "litellm-model_prices_and_context_window.json"),
-    metaPath: import_node_path4.default.join(cacheDir, "litellm-model_prices_and_context_window.meta.json")
+    catalogPath: import_node_path5.default.join(cacheDir, "litellm-model_prices_and_context_window.json"),
+    metaPath: import_node_path5.default.join(cacheDir, "litellm-model_prices_and_context_window.meta.json")
   };
 }
 async function readJsonFile(filePath) {
   try {
-    const raw = await import_promises3.default.readFile(filePath, "utf8");
+    const raw = await import_promises4.default.readFile(filePath, "utf8");
     return JSON.parse(raw);
   } catch {
     return null;
   }
 }
 async function writeJsonFile(filePath, value) {
-  await import_promises3.default.mkdir(import_node_path4.default.dirname(filePath), { recursive: true });
-  await import_promises3.default.writeFile(filePath, `${JSON.stringify(value, null, 2)}
+  await import_promises4.default.mkdir(import_node_path5.default.dirname(filePath), { recursive: true });
+  await import_promises4.default.writeFile(filePath, `${JSON.stringify(value, null, 2)}
 `, "utf8");
 }
 function isStale(meta3, nowMs) {
@@ -75509,7 +75588,7 @@ function withDefaultCacheDir(env3) {
   }
   const home = env3.HOME?.trim();
   if (!home) return env3;
-  return { ...env3, TOKENTALLY_CACHE_DIR: import_node_path5.default.join(home, ".summarize", "cache") };
+  return { ...env3, TOKENTALLY_CACHE_DIR: import_node_path6.default.join(home, ".summarize", "cache") };
 }
 async function loadLiteLlmCatalog2({
   env: env3,
@@ -75554,6 +75633,27 @@ ${headerLines.length > 0 ? `${headerLines.join("\n")}
 ` : ""}Return only the summary.`;
   return prompt;
 }
+function buildFileTextSummaryPrompt({
+  filename,
+  originalMediaType,
+  contentMediaType,
+  summaryLength,
+  contentLength
+}) {
+  const effectiveSummaryLength = typeof summaryLength === "string" ? summaryLength : summaryLength.maxCharacters > contentLength ? { maxCharacters: contentLength } : summaryLength;
+  const maxCharactersLine = typeof effectiveSummaryLength === "string" ? "" : `Target length: around ${effectiveSummaryLength.maxCharacters.toLocaleString()} characters total (including Markdown and whitespace). This is a soft guideline; prioritize clarity.`;
+  const headerLines = [
+    filename ? `Filename: ${filename}` : null,
+    originalMediaType ? `Original media type: ${originalMediaType}` : null,
+    `Provided as: ${contentMediaType}`,
+    `Extracted content length: ${contentLength.toLocaleString()} characters. Do not exceed the extracted content length; if the requested length is larger, keep the summary at or below the extracted content length and do not add details.`
+  ].filter(Boolean);
+  return `You summarize files for curious users. Summarize the file content below. Be factual and do not invent details. Format the answer in Markdown. Do not use emojis. ${maxCharactersLine}
+
+${headerLines.length > 0 ? `${headerLines.join("\n")}
+
+` : ""}Return only the summary.`;
+}
 
 // src/prompts/link-summary.ts
 var SUMMARY_LENGTH_DIRECTIVES = {
@@ -78283,16 +78383,16 @@ function startSpinner({
 
 // src/version.ts
 var import_node_fs5 = __toESM(require("node:fs"), 1);
-var import_node_path6 = __toESM(require("node:path"), 1);
+var import_node_path7 = __toESM(require("node:path"), 1);
 var import_node_url = require("node:url");
-var FALLBACK_VERSION = "0.3.0";
+var FALLBACK_VERSION = "0.4.0";
 function resolvePackageVersion(importMetaUrl) {
   const injected = typeof process !== "undefined" && typeof process.env.SUMMARIZE_VERSION === "string" ? process.env.SUMMARIZE_VERSION.trim() : "";
   if (injected.length > 0) return injected;
   const startDir = (() => {
     if (typeof importMetaUrl === "string" && importMetaUrl.trim().length > 0) {
       try {
-        return import_node_path6.default.dirname((0, import_node_url.fileURLToPath)(importMetaUrl));
+        return import_node_path7.default.dirname((0, import_node_url.fileURLToPath)(importMetaUrl));
       } catch {
       }
     }
@@ -78301,7 +78401,7 @@ function resolvePackageVersion(importMetaUrl) {
   })();
   let dir = startDir;
   for (let i = 0; i < 10; i += 1) {
-    const candidate = import_node_path6.default.join(dir, "package.json");
+    const candidate = import_node_path7.default.join(dir, "package.json");
     try {
       const raw = import_node_fs5.default.readFileSync(candidate, "utf8");
       const json3 = JSON.parse(raw);
@@ -78310,7 +78410,7 @@ function resolvePackageVersion(importMetaUrl) {
       }
     } catch {
     }
-    const parent = import_node_path6.default.dirname(dir);
+    const parent = import_node_path7.default.dirname(dir);
     if (parent === dir) break;
     dir = parent;
   }
@@ -78319,6 +78419,7 @@ function resolvePackageVersion(importMetaUrl) {
 
 // src/run.ts
 var BIRD_TIP = "Tip: Install bird\u{1F426} for better Twitter support: https://github.com/steipete/bird";
+var UVX_TIP = "Tip: Install uv (uvx) for local Markdown conversion: brew install uv (or set UVX_PATH to your uvx binary).";
 var TWITTER_HOSTS2 = /* @__PURE__ */ new Set(["x.com", "twitter.com", "mobile.twitter.com"]);
 var SUMMARY_LENGTH_MAX_CHARACTERS = {
   short: 1200,
@@ -78350,10 +78451,22 @@ function isExecutable(filePath) {
 }
 function hasBirdCli(env3) {
   const candidates = [];
-  const pathEnv = env3.PATH ?? process.env.PATH ?? "";
-  for (const entry of pathEnv.split(import_node_path7.default.delimiter)) {
+  const pathEnv = env3.PATH ?? "";
+  for (const entry of pathEnv.split(import_node_path8.default.delimiter)) {
     if (!entry) continue;
-    candidates.push(import_node_path7.default.join(entry, "bird"));
+    candidates.push(import_node_path8.default.join(entry, "bird"));
+  }
+  return candidates.some((candidate) => isExecutable(candidate));
+}
+function hasUvxCli(env3) {
+  if (typeof env3.UVX_PATH === "string" && env3.UVX_PATH.trim().length > 0) {
+    return true;
+  }
+  const candidates = [];
+  const pathEnv = env3.PATH ?? "";
+  for (const entry of pathEnv.split(import_node_path8.default.delimiter)) {
+    if (!entry) continue;
+    candidates.push(import_node_path8.default.join(entry, "uvx"));
   }
   return candidates.some((candidate) => isExecutable(candidate));
 }
@@ -78404,6 +78517,15 @@ function withBirdTip(error47, url2, env3) {
 ${BIRD_TIP}`;
   return error47 instanceof Error ? new Error(combined, { cause: error47 }) : new Error(combined);
 }
+function withUvxTip(error47, env3) {
+  if (hasUvxCli(env3)) {
+    return error47 instanceof Error ? error47 : new Error(String(error47));
+  }
+  const message = error47 instanceof Error ? error47.message : String(error47);
+  const combined = `${message}
+${UVX_TIP}`;
+  return error47 instanceof Error ? new Error(combined, { cause: error47 }) : new Error(combined);
+}
 var MAX_TEXT_BYTES_DEFAULT = 10 * 1024 * 1024;
 function buildProgram() {
   return new Command().name("summarize").description("Summarize web pages and YouTube links (uses direct provider API keys).").argument("[input]", "URL or local file path to summarize").option(
@@ -78412,12 +78534,27 @@ function buildProgram() {
     "auto"
   ).option(
     "--firecrawl <mode>",
-    "Firecrawl usage: off, auto (fallback), always (try Firecrawl first).",
+    "Firecrawl usage: off, auto (fallback), always (try Firecrawl first). Note: in --format md website mode, defaults to always when FIRECRAWL_API_KEY is set (unless --firecrawl is set explicitly).",
     "auto"
   ).option(
-    "--markdown <mode>",
-    "Website Markdown output: off, auto (use LLM when configured), llm (force LLM). Only affects --extract-only for non-YouTube URLs.",
-    "auto"
+    "--format <format>",
+    "Website/file content format: md|text. For websites: controls the extraction format. For files: controls whether we try to preprocess to Markdown for model compatibility. (default: text)",
+    "text"
+  ).addOption(
+    new Option(
+      "--preprocess <mode>",
+      "Preprocess inputs for model compatibility: off, auto (fallback), always."
+    ).choices(["off", "auto", "always"]).default("auto")
+  ).addOption(
+    new Option(
+      "--markdown-mode <mode>",
+      "HTML\u2192Markdown conversion: off, auto (prefer Firecrawl when configured, then LLM when configured, then markitdown when available), llm (force LLM). Only affects --format md for non-YouTube URLs."
+    ).default("auto")
+  ).addOption(
+    new Option(
+      "--markdown <mode>",
+      "Deprecated alias for --markdown-mode (use --extract --format md --markdown-mode ...)"
+    ).hideHelp()
   ).option(
     "--length <length>",
     "Summary length: short|medium|long|xl|xxl or a character limit like 20000, 20k",
@@ -78434,7 +78571,7 @@ function buildProgram() {
     "--model <model>",
     "LLM model id (gateway-style): xai/..., openai/..., google/... (default: google/gemini-3-flash-preview)",
     void 0
-  ).option("--extract-only", "Print extracted content and exit (no LLM summary)", false).option("--json", "Output structured JSON (includes prompt + metrics)", false).option(
+  ).option("--extract", "Print extracted content and exit (no LLM summary)", false).addOption(new Option("--extract-only", "Deprecated alias for --extract").hideHelp()).option("--json", "Output structured JSON (includes prompt + metrics)", false).option(
     "--stream <mode>",
     "Stream LLM output: auto (TTY only), on, off. Note: streaming is disabled in --json mode.",
     "auto"
@@ -78550,6 +78687,22 @@ function getTextContentFromAttachment(attachment) {
   }
   return { content: "", bytes: 0 };
 }
+function getFileBytesFromAttachment(attachment) {
+  if (attachment.part.type !== "file") return null;
+  const data = attachment.part.data;
+  return data instanceof Uint8Array ? data : null;
+}
+function shouldMarkitdownConvertMediaType(mediaType) {
+  const mt = mediaType.toLowerCase();
+  if (mt === "application/pdf") return true;
+  if (mt === "application/rtf") return true;
+  if (mt === "text/html" || mt === "application/xhtml+xml") return true;
+  if (mt === "application/msword") return true;
+  if (mt.startsWith("application/vnd.openxmlformats-officedocument.")) return true;
+  if (mt === "application/vnd.ms-excel") return true;
+  if (mt === "application/vnd.ms-powerpoint") return true;
+  return false;
+}
 function assertProviderSupportsAttachment({
   provider,
   modelId,
@@ -78613,9 +78766,10 @@ function attachRichHelp(program2, env3, stdout) {
     () => `
 ${heading("Examples")}
   ${cmd('summarize "https://example.com"')}
-  ${cmd('summarize "https://example.com" --extract-only')} ${dim3("# website markdown (LLM if configured)")}
-  ${cmd('summarize "https://example.com" --extract-only --markdown llm')} ${dim3("# website markdown via LLM")}
-  ${cmd('summarize "https://www.youtube.com/watch?v=I845O57ZSy4&t=11s" --extract-only --youtube web')}
+  ${cmd('summarize "https://example.com" --extract')} ${dim3("# extracted plain text")}
+  ${cmd('summarize "https://example.com" --extract --format md')} ${dim3("# extracted markdown (prefers Firecrawl when configured)")}
+  ${cmd('summarize "https://example.com" --extract --format md --markdown-mode llm')} ${dim3("# extracted markdown via LLM")}
+  ${cmd('summarize "https://www.youtube.com/watch?v=I845O57ZSy4&t=11s" --extract --youtube web')}
   ${cmd('summarize "https://example.com" --length 20k --max-output-tokens 2k --timeout 2m --model openai/gpt-5.2')}
   ${cmd('OPENROUTER_API_KEY=... summarize "https://example.com" --model openai/openai/gpt-oss-20b')}
   ${cmd('summarize "https://example.com" --json --verbose')}
@@ -78756,10 +78910,11 @@ function writeFinishLine({
   stderr.write(`${ansi("1;32", line, color2)}
 `);
 }
-async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
+async function runCli(argv, { env: env3, fetch: fetch2, execFile: execFileOverride, stdout, stderr }) {
   ;
   globalThis.AI_SDK_LOG_WARNINGS = false;
   const normalizedArgv = argv.filter((arg) => arg !== "--");
+  const execFileImpl = execFileOverride ?? import_node_child_process2.execFile;
   const version2 = resolvePackageVersion();
   const program2 = buildProgram();
   program2.configureOutput({
@@ -78800,7 +78955,7 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
     program2.opts().maxOutputTokens
   );
   const timeoutMs = parseDurationMs(program2.opts().timeout);
-  const extractOnly = Boolean(program2.opts().extractOnly);
+  const extractMode = Boolean(program2.opts().extract) || Boolean(program2.opts().extractOnly);
   const json3 = Boolean(program2.opts().json);
   const streamMode = parseStreamMode(program2.opts().stream);
   const renderMode = parseRenderMode(program2.opts().render);
@@ -78808,9 +78963,19 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
   const metricsMode = parseMetricsMode(program2.opts().metrics);
   const metricsEnabled = metricsMode !== "off";
   const metricsDetailed = metricsMode === "detailed";
-  const markdownMode = parseMarkdownMode(program2.opts().markdown);
+  const preprocessMode = parsePreprocessMode(program2.opts().preprocess);
+  const format2 = parseExtractFormat(program2.opts().format);
   const shouldComputeReport = metricsEnabled;
   const isYoutubeUrl = typeof url2 === "string" ? /youtube\.com|youtu\.be/i.test(url2) : false;
+  const firecrawlExplicitlySet = normalizedArgv.some(
+    (arg) => arg === "--firecrawl" || arg.startsWith("--firecrawl=")
+  );
+  const markdownModeExplicitlySet = normalizedArgv.some(
+    (arg) => arg === "--markdown-mode" || arg.startsWith("--markdown-mode=") || arg === "--markdown" || arg.startsWith("--markdown=")
+  );
+  const markdownMode = format2 === "markdown" ? parseMarkdownMode(
+    program2.opts().markdownMode ?? program2.opts().markdown ?? "auto"
+  ) : "off";
   const requestedFirecrawlMode = parseFirecrawlMode(program2.opts().firecrawl);
   const modelArg = typeof program2.opts().model === "string" ? program2.opts().model : null;
   const { config: config2, path: configPath } = loadSummarizeConfig({ env: env3 });
@@ -78839,6 +79004,12 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
   const anthropicConfigured = typeof anthropicApiKey === "string" && anthropicApiKey.length > 0;
   const openrouterConfigured = typeof openrouterApiKey === "string" && openrouterApiKey.length > 0;
   const openrouterOptions = openRouterProviders ? { providers: openRouterProviders } : void 0;
+  if (markdownModeExplicitlySet && format2 !== "markdown") {
+    throw new Error("--markdown-mode is only supported with --format md");
+  }
+  if (markdownModeExplicitlySet && inputTarget.kind !== "url") {
+    throw new Error("--markdown-mode is only supported for website URLs");
+  }
   const llmCalls = [];
   let firecrawlRequests = 0;
   let apifyRequests = 0;
@@ -78932,7 +79103,7 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
     if (streamMode !== "auto") return streamMode;
     return isRichTty(stdout) ? "on" : "off";
   })();
-  const streamingEnabled = effectiveStreamMode === "on" && !json3 && !extractOnly;
+  const streamingEnabled = effectiveStreamMode === "on" && !json3 && !extractMode;
   const effectiveRenderMode = (() => {
     if (renderMode !== "auto") return renderMode;
     if (!isRichTty(stdout)) return "plain";
@@ -78957,8 +79128,8 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
 `
     );
   };
-  if (extractOnly && inputTarget.kind !== "url") {
-    throw new Error("--extract-only is only supported for website/YouTube URLs");
+  if (extractMode && inputTarget.kind !== "url") {
+    throw new Error("--extract is only supported for website/YouTube URLs");
   }
   const progressEnabled = isRichTty(stderr) && !verbose && !json3;
   let clearProgressBeforeStdout = null;
@@ -78987,11 +79158,6 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
         `Missing ${requiredKeyEnv} for model ${parsedModel.canonical}. Set the env var or choose a different --model.`
       );
     }
-    assertProviderSupportsAttachment({
-      provider: parsedModel.provider,
-      modelId: parsedModel.canonical,
-      attachment: { part: attachment.part, mediaType: attachment.mediaType }
-    });
     const modelResolution = await resolveModelIdForLlmCall({
       parsedModel,
       apiKeys: { googleApiKey: apiKeysForLlm.googleApiKey },
@@ -79013,14 +79179,114 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
         `Text file too large (${formatBytes(textContent.bytes)}). Limit is ${formatBytes(MAX_TEXT_BYTES_DEFAULT)}.`
       );
     }
+    const fileBytes = getFileBytesFromAttachment(attachment);
+    const canPreprocessWithMarkitdown = format2 === "markdown" && preprocessMode !== "off" && hasUvxCli(env3) && attachment.part.type === "file" && fileBytes !== null && shouldMarkitdownConvertMediaType(attachment.mediaType);
     const summaryLengthTarget = lengthArg.kind === "preset" ? lengthArg.preset : { maxCharacters: lengthArg.maxCharacters };
-    const promptText = buildFileSummaryPrompt({
-      filename: attachment.filename,
-      mediaType: attachment.mediaType,
-      summaryLength: summaryLengthTarget,
-      contentLength: textContent?.content.length ?? null
-    });
-    const promptPayload = buildAssetPromptPayload({ promptText, attachment, textContent });
+    let promptText = "";
+    const buildAttachmentPromptPayload = () => {
+      promptText = buildFileSummaryPrompt({
+        filename: attachment.filename,
+        mediaType: attachment.mediaType,
+        summaryLength: summaryLengthTarget,
+        contentLength: textContent?.content.length ?? null
+      });
+      return buildAssetPromptPayload({ promptText, attachment, textContent });
+    };
+    const buildMarkitdownPromptPayload = (markdown) => {
+      promptText = buildFileTextSummaryPrompt({
+        filename: attachment.filename,
+        originalMediaType: attachment.mediaType,
+        contentMediaType: "text/markdown",
+        summaryLength: summaryLengthTarget,
+        contentLength: markdown.length
+      });
+      return `${promptText}
+
+---
+
+${markdown}`.trim();
+    };
+    let preprocessedMarkdown = null;
+    let usingPreprocessedMarkdown = false;
+    if (preprocessMode === "always" && canPreprocessWithMarkitdown) {
+      if (!fileBytes) {
+        throw new Error("Internal error: missing file bytes for markitdown preprocessing");
+      }
+      try {
+        preprocessedMarkdown = await convertToMarkdownWithMarkitdown({
+          bytes: fileBytes,
+          filenameHint: attachment.filename,
+          mediaTypeHint: attachment.mediaType,
+          uvxCommand: env3.UVX_PATH,
+          timeoutMs,
+          env: env3,
+          execFileImpl
+        });
+      } catch (error47) {
+        const message = error47 instanceof Error ? error47.message : String(error47);
+        throw new Error(
+          `Failed to preprocess ${attachment.mediaType} with markitdown: ${message} (disable with --preprocess off).`
+        );
+      }
+      if (Buffer.byteLength(preprocessedMarkdown, "utf8") > MAX_TEXT_BYTES_DEFAULT) {
+        throw new Error(
+          `Preprocessed Markdown too large (${formatBytes(Buffer.byteLength(preprocessedMarkdown, "utf8"))}). Limit is ${formatBytes(MAX_TEXT_BYTES_DEFAULT)}.`
+        );
+      }
+      usingPreprocessedMarkdown = true;
+    }
+    let promptPayload = buildAttachmentPromptPayload();
+    if (usingPreprocessedMarkdown) {
+      if (!preprocessedMarkdown) {
+        throw new Error("Internal error: missing markitdown content for preprocessing");
+      }
+      promptPayload = buildMarkitdownPromptPayload(preprocessedMarkdown);
+    }
+    if (!usingPreprocessedMarkdown) {
+      try {
+        assertProviderSupportsAttachment({
+          provider: parsedModel.provider,
+          modelId: parsedModel.canonical,
+          attachment: { part: attachment.part, mediaType: attachment.mediaType }
+        });
+      } catch (error47) {
+        if (!canPreprocessWithMarkitdown) {
+          if (format2 === "markdown" && preprocessMode !== "off" && attachment.part.type === "file" && shouldMarkitdownConvertMediaType(attachment.mediaType) && !hasUvxCli(env3)) {
+            throw withUvxTip(error47, env3);
+          }
+          throw error47;
+        }
+        if (!fileBytes) {
+          throw new Error("Internal error: missing file bytes for markitdown preprocessing");
+        }
+        try {
+          preprocessedMarkdown = await convertToMarkdownWithMarkitdown({
+            bytes: fileBytes,
+            filenameHint: attachment.filename,
+            mediaTypeHint: attachment.mediaType,
+            uvxCommand: env3.UVX_PATH,
+            timeoutMs,
+            env: env3,
+            execFileImpl
+          });
+        } catch (markitdownError) {
+          if (preprocessMode === "auto") {
+            throw error47;
+          }
+          const message = markitdownError instanceof Error ? markitdownError.message : String(markitdownError);
+          throw new Error(
+            `Failed to preprocess ${attachment.mediaType} with markitdown: ${message} (disable with --preprocess off).`
+          );
+        }
+        if (Buffer.byteLength(preprocessedMarkdown, "utf8") > MAX_TEXT_BYTES_DEFAULT) {
+          throw new Error(
+            `Preprocessed Markdown too large (${formatBytes(Buffer.byteLength(preprocessedMarkdown, "utf8"))}). Limit is ${formatBytes(MAX_TEXT_BYTES_DEFAULT)}.`
+          );
+        }
+        usingPreprocessedMarkdown = true;
+        promptPayload = buildMarkitdownPromptPayload(preprocessedMarkdown);
+      }
+    }
     const maxInputTokensForCall = await resolveMaxInputTokensForCall(parsedModelEffective.canonical);
     if (typeof maxInputTokensForCall === "number" && Number.isFinite(maxInputTokensForCall) && maxInputTokensForCall > 0 && typeof promptPayload === "string") {
       const tokenCount = countTokens(promptPayload);
@@ -79309,7 +79575,7 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
   if (inputTarget.kind === "file") {
     let sizeLabel = null;
     try {
-      const stat = await import_promises4.default.stat(inputTarget.filePath);
+      const stat = await import_promises5.default.stat(inputTarget.filePath);
       if (stat.isFile()) {
         sizeLabel = formatBytes(stat.size);
       }
@@ -79411,22 +79677,33 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
   if (!url2) {
     throw new Error("Only HTTP and HTTPS URLs can be summarized");
   }
-  const firecrawlMode = requestedFirecrawlMode;
+  const wantsMarkdown = format2 === "markdown" && !isYoutubeUrl;
+  if (wantsMarkdown && markdownMode === "off") {
+    throw new Error("--format md conflicts with --markdown-mode off (use --format text)");
+  }
+  const firecrawlMode = (() => {
+    if (wantsMarkdown && !isYoutubeUrl && !firecrawlExplicitlySet && firecrawlConfigured) {
+      return "always";
+    }
+    return requestedFirecrawlMode;
+  })();
   if (firecrawlMode === "always" && !firecrawlConfigured) {
     throw new Error("--firecrawl always requires FIRECRAWL_API_KEY");
   }
-  const effectiveMarkdownMode = markdownMode;
-  const markdownRequested = extractOnly && !isYoutubeUrl && effectiveMarkdownMode !== "off";
+  const markdownRequested = wantsMarkdown;
+  const effectiveMarkdownMode = markdownRequested ? markdownMode : "off";
   const hasKeyForModel = parsedModelForLlm.provider === "xai" ? xaiConfigured : parsedModelForLlm.provider === "google" ? googleConfigured : parsedModelForLlm.provider === "anthropic" ? anthropicConfigured : Boolean(apiKey);
   const markdownProvider = hasKeyForModel ? parsedModelForLlm.provider : "none";
   if (markdownRequested && effectiveMarkdownMode === "llm" && !hasKeyForModel) {
     const required2 = parsedModelForLlm.provider === "xai" ? "XAI_API_KEY" : parsedModelForLlm.provider === "google" ? "GEMINI_API_KEY (or GOOGLE_GENERATIVE_AI_API_KEY / GOOGLE_API_KEY)" : parsedModelForLlm.provider === "anthropic" ? "ANTHROPIC_API_KEY" : "OPENAI_API_KEY";
-    throw new Error(`--markdown llm requires ${required2} for model ${parsedModelForLlm.canonical}`);
+    throw new Error(
+      `--markdown-mode llm requires ${required2} for model ${parsedModelForLlm.canonical}`
+    );
   }
   writeVerbose(
     stderr,
     verbose,
-    `config url=${url2} timeoutMs=${timeoutMs} youtube=${youtubeMode} firecrawl=${firecrawlMode} length=${lengthArg.kind === "preset" ? lengthArg.preset : `${lengthArg.maxCharacters} chars`} maxOutputTokens=${formatOptionalNumber(maxOutputTokensArg)} json=${json3} extractOnly=${extractOnly} markdown=${effectiveMarkdownMode} model=${model} stream=${effectiveStreamMode} render=${effectiveRenderMode}`,
+    `config url=${url2} timeoutMs=${timeoutMs} youtube=${youtubeMode} firecrawl=${firecrawlMode} length=${lengthArg.kind === "preset" ? lengthArg.preset : `${lengthArg.maxCharacters} chars`} maxOutputTokens=${formatOptionalNumber(maxOutputTokensArg)} json=${json3} extract=${extractMode} format=${format2} preprocess=${preprocessMode} markdownMode=${markdownMode} model=${model} stream=${effectiveStreamMode} render=${effectiveRenderMode}`,
     verboseColor
   );
   writeVerbose(
@@ -79450,7 +79727,7 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
     verboseColor
   );
   const scrapeWithFirecrawl = firecrawlConfigured && firecrawlMode !== "off" ? createFirecrawlScraper({ apiKey: firecrawlApiKey, fetchImpl: trackedFetch }) : null;
-  const convertHtmlToMarkdown = markdownRequested && (effectiveMarkdownMode === "llm" || markdownProvider !== "none") ? createHtmlToMarkdownConverter({
+  const llmHtmlToMarkdown = markdownRequested && (effectiveMarkdownMode === "llm" || markdownProvider !== "none") ? createHtmlToMarkdownConverter({
     modelId: model,
     xaiApiKey: xaiConfigured ? xaiApiKey : null,
     googleApiKey: googleConfigured ? googleApiKey : null,
@@ -79463,6 +79740,40 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
       llmCalls.push({ provider, model: usedModel, usage, purpose: "markdown" });
     }
   }) : null;
+  const markitdownHtmlToMarkdown = markdownRequested && preprocessMode !== "off" && hasUvxCli(env3) ? async (args) => {
+    void args.url;
+    void args.title;
+    void args.siteName;
+    return convertToMarkdownWithMarkitdown({
+      bytes: new TextEncoder().encode(args.html),
+      filenameHint: "page.html",
+      mediaTypeHint: "text/html",
+      uvxCommand: env3.UVX_PATH,
+      timeoutMs: args.timeoutMs,
+      env: env3,
+      execFileImpl
+    });
+  } : null;
+  const convertHtmlToMarkdown = markdownRequested ? async (args) => {
+    if (effectiveMarkdownMode === "llm") {
+      if (!llmHtmlToMarkdown) {
+        throw new Error("No HTML\u2192Markdown converter configured");
+      }
+      return llmHtmlToMarkdown(args);
+    }
+    if (llmHtmlToMarkdown) {
+      try {
+        return await llmHtmlToMarkdown(args);
+      } catch (error47) {
+        if (!markitdownHtmlToMarkdown) throw error47;
+        return await markitdownHtmlToMarkdown(args);
+      }
+    }
+    if (markitdownHtmlToMarkdown) {
+      return await markitdownHtmlToMarkdown(args);
+    }
+    throw new Error("No HTML\u2192Markdown converter configured");
+  } : null;
   const readTweetWithBirdClient = hasBirdCli(env3) ? ({ url: url3, timeoutMs: timeoutMs2 }) => readTweetWithBird({ url: url3, timeoutMs: timeoutMs2, env: env3 }) : null;
   writeVerbose(stderr, verbose, "extract start", verboseColor);
   const stopOscProgress = startOscProgress2({
@@ -79643,7 +79954,7 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
     if (progressEnabled) {
       websiteProgress?.stop?.();
       spinner.setText(
-        extractOnly ? `Extracted (${extractedContentSize}${viaSourceLabel})` : `Summarizing (sent ${extractedContentSize}${viaSourceLabel})\u2026`
+        extractMode ? `Extracted (${extractedContentSize}${viaSourceLabel})` : `Summarizing (sent ${extractedContentSize}${viaSourceLabel})\u2026`
       );
     }
     writeVerbose(
@@ -79688,6 +79999,10 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
       )} attemptedProviders=${extracted.diagnostics.transcript.attemptedProviders.length > 0 ? extracted.diagnostics.transcript.attemptedProviders.join(",") : "none"} notes=${formatOptionalString(extracted.diagnostics.transcript.notes ?? null)}`,
       verboseColor
     );
+    if (extractMode && markdownRequested && preprocessMode !== "off" && effectiveMarkdownMode === "auto" && !extracted.diagnostics.markdown.used && !hasUvxCli(env3)) {
+      stderr.write(`${UVX_TIP}
+`);
+    }
     const isYouTube = extracted.siteName === "YouTube";
     const prompt = buildLinkSummaryPrompt({
       url: extracted.url,
@@ -79700,7 +80015,7 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
       summaryLength: lengthArg.kind === "preset" ? lengthArg.preset : { maxCharacters: lengthArg.maxCharacters },
       shares: []
     });
-    if (extractOnly) {
+    if (extractMode) {
       clearProgressForStdout();
       if (json3) {
         const finishReport = shouldComputeReport ? await buildReport() : null;
@@ -79711,6 +80026,7 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
             timeoutMs,
             youtube: youtubeMode,
             firecrawl: firecrawlMode,
+            format: format2,
             markdown: effectiveMarkdownMode,
             length: lengthArg.kind === "preset" ? { kind: "preset", preset: lengthArg.preset } : { kind: "chars", maxCharacters: lengthArg.maxCharacters },
             maxOutputTokens: maxOutputTokensArg,
@@ -79783,6 +80099,7 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
             timeoutMs,
             youtube: youtubeMode,
             firecrawl: firecrawlMode,
+            format: format2,
             markdown: effectiveMarkdownMode,
             length: lengthArg.kind === "preset" ? { kind: "preset", preset: lengthArg.preset } : { kind: "chars", maxCharacters: lengthArg.maxCharacters },
             maxOutputTokens: maxOutputTokensArg,
@@ -80057,6 +80374,7 @@ async function runCli(argv, { env: env3, fetch: fetch2, stdout, stderr }) {
           timeoutMs,
           youtube: youtubeMode,
           firecrawl: firecrawlMode,
+          format: format2,
           markdown: effectiveMarkdownMode,
           length: lengthArg.kind === "preset" ? { kind: "preset", preset: lengthArg.preset } : { kind: "chars", maxCharacters: lengthArg.maxCharacters },
           maxOutputTokens: maxOutputTokensArg,