@steipete/summarize 0.3.0 → 0.4.0

package/dist/esm/run.js

@@ -11,17 +11,19 @@ import { buildAssetPromptMessages, classifyUrl, loadLocalAsset, loadRemoteAsset,
 import { createLinkPreviewClient } from './content/index.js';
 import { buildRunMetricsReport } from './costs.js';
 import { createFirecrawlScraper } from './firecrawl.js';
-import { parseDurationMs, parseFirecrawlMode, parseLengthArg, parseMarkdownMode, parseMaxOutputTokensArg, parseMetricsMode, parseRenderMode, parseStreamMode, parseYoutubeMode, } from './flags.js';
+import { parseDurationMs, parseExtractFormat, parseFirecrawlMode, parseLengthArg, parseMarkdownMode, parseMaxOutputTokensArg, parseMetricsMode, parsePreprocessMode, parseRenderMode, parseStreamMode, parseYoutubeMode, } from './flags.js';
 import { generateTextWithModelId, streamTextWithModelId } from './llm/generate-text.js';
 import { resolveGoogleModelForUsage } from './llm/google-models.js';
 import { createHtmlToMarkdownConverter } from './llm/html-to-markdown.js';
 import { normalizeGatewayStyleModelId, parseGatewayStyleModelId } from './llm/model-id.js';
+import { convertToMarkdownWithMarkitdown } from './markitdown.js';
 import { loadLiteLlmCatalog, resolveLiteLlmMaxInputTokensForModelId, resolveLiteLlmMaxOutputTokensForModelId, resolveLiteLlmPricingForModelId, } from './pricing/litellm.js';
-import { buildFileSummaryPrompt, buildLinkSummaryPrompt } from './prompts/index.js';
+import { buildFileSummaryPrompt, buildFileTextSummaryPrompt, buildLinkSummaryPrompt, } from './prompts/index.js';
 import { startOscProgress } from './tty/osc-progress.js';
 import { startSpinner } from './tty/spinner.js';
 import { resolvePackageVersion } from './version.js';
 const BIRD_TIP = 'Tip: Install bird🐦 for better Twitter support: https://github.com/steipete/bird';
+const UVX_TIP = 'Tip: Install uv (uvx) for local Markdown conversion: brew install uv (or set UVX_PATH to your uvx binary).';
 const TWITTER_HOSTS = new Set(['x.com', 'twitter.com', 'mobile.twitter.com']);
 const SUMMARY_LENGTH_MAX_CHARACTERS = {
     short: 1200,
@@ -58,7 +60,7 @@ function isExecutable(filePath) {
 }
 function hasBirdCli(env) {
     const candidates = [];
-    const pathEnv = env.PATH ?? process.env.PATH ?? '';
+    const pathEnv = env.PATH ?? '';
     for (const entry of pathEnv.split(path.delimiter)) {
         if (!entry)
             continue;
@@ -66,6 +68,19 @@ function hasBirdCli(env) {
     }
     return candidates.some((candidate) => isExecutable(candidate));
 }
+function hasUvxCli(env) {
+    if (typeof env.UVX_PATH === 'string' && env.UVX_PATH.trim().length > 0) {
+        return true;
+    }
+    const candidates = [];
+    const pathEnv = env.PATH ?? '';
+    for (const entry of pathEnv.split(path.delimiter)) {
+        if (!entry)
+            continue;
+        candidates.push(path.join(entry, 'uvx'));
+    }
+    return candidates.some((candidate) => isExecutable(candidate));
+}
 async function readTweetWithBird(args) {
     return await new Promise((resolve, reject) => {
         execFile('bird', ['read', args.url, '--json'], {
@@ -108,6 +123,14 @@ function withBirdTip(error, url, env) {
     const combined = `${message}\n${BIRD_TIP}`;
     return error instanceof Error ? new Error(combined, { cause: error }) : new Error(combined);
 }
+function withUvxTip(error, env) {
+    if (hasUvxCli(env)) {
+        return error instanceof Error ? error : new Error(String(error));
+    }
+    const message = error instanceof Error ? error.message : String(error);
+    const combined = `${message}\n${UVX_TIP}`;
+    return error instanceof Error ? new Error(combined, { cause: error }) : new Error(combined);
+}
 const MAX_TEXT_BYTES_DEFAULT = 10 * 1024 * 1024;
 function buildProgram() {
     return new Command()
@@ -115,13 +138,19 @@ function buildProgram() {
         .description('Summarize web pages and YouTube links (uses direct provider API keys).')
         .argument('[input]', 'URL or local file path to summarize')
         .option('--youtube <mode>', 'YouTube transcript source: auto, web (youtubei/captionTracks), yt-dlp (audio+whisper), apify', 'auto')
-        .option('--firecrawl <mode>', 'Firecrawl usage: off, auto (fallback), always (try Firecrawl first).', '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')
+        .option('--firecrawl <mode>', '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('--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→Markdown 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', 'medium')
         .option('--max-output-tokens <count>', 'Hard cap for LLM output tokens (e.g. 2000, 2k). Overrides provider defaults.', undefined)
         .option('--timeout <duration>', 'Timeout for content fetching and LLM request: 30 (seconds), 30s, 2m, 5000ms', '2m')
         .option('--model <model>', 'LLM model id (gateway-style): xai/..., openai/..., google/... (default: google/gemini-3-flash-preview)', undefined)
-        .option('--extract-only', 'Print extracted content and exit (no LLM summary)', false)
+        .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')
         .option('--render <mode>', 'Render Markdown output: auto (TTY only), md-live, md, plain. Note: auto selects md-live when streaming to a TTY.', 'auto')
@@ -250,6 +279,30 @@ 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, attachment, }) {
     // xAI via AI SDK currently supports image parts, but not generic file parts (e.g. PDFs).
     if (provider === 'xai' &&
@@ -314,9 +367,10 @@ function attachRichHelp(program, env, stdout) {
     program.addHelpText('after', () => `
 ${heading('Examples')}
   ${cmd('summarize "https://example.com"')}
-  ${cmd('summarize "https://example.com" --extract-only')} ${dim('# website markdown (LLM if configured)')}
-  ${cmd('summarize "https://example.com" --extract-only --markdown llm')} ${dim('# 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')} ${dim('# extracted plain text')}
+  ${cmd('summarize "https://example.com" --extract --format md')} ${dim('# extracted markdown (prefers Firecrawl when configured)')}
+  ${cmd('summarize "https://example.com" --extract --format md --markdown-mode llm')} ${dim('# 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')}
   ${cmd('summarize "https://example.com" --json --verbose')}
@@ -448,10 +502,11 @@ function writeFinishLine({ stderr, elapsedMs, model, report, costUsd, color, })
     stderr.write('\n');
     stderr.write(`${ansi('1;32', line, color)}\n`);
 }
-export async function runCli(argv, { env, fetch, stdout, stderr }) {
+export async function runCli(argv, { env, fetch, execFile: execFileOverride, stdout, stderr }) {
     ;
     globalThis.AI_SDK_LOG_WARNINGS = false;
     const normalizedArgv = argv.filter((arg) => arg !== '--');
+    const execFileImpl = execFileOverride ?? execFile;
     const version = resolvePackageVersion();
     const program = buildProgram();
     program.configureOutput({
@@ -488,7 +543,7 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
     const lengthArg = parseLengthArg(program.opts().length);
     const maxOutputTokensArg = parseMaxOutputTokensArg(program.opts().maxOutputTokens);
     const timeoutMs = parseDurationMs(program.opts().timeout);
-    const extractOnly = Boolean(program.opts().extractOnly);
+    const extractMode = Boolean(program.opts().extract) || Boolean(program.opts().extractOnly);
     const json = Boolean(program.opts().json);
     const streamMode = parseStreamMode(program.opts().stream);
     const renderMode = parseRenderMode(program.opts().render);
@@ -496,9 +551,20 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
     const metricsMode = parseMetricsMode(program.opts().metrics);
     const metricsEnabled = metricsMode !== 'off';
     const metricsDetailed = metricsMode === 'detailed';
-    const markdownMode = parseMarkdownMode(program.opts().markdown);
+    const preprocessMode = parsePreprocessMode(program.opts().preprocess);
+    const format = parseExtractFormat(program.opts().format);
     const shouldComputeReport = metricsEnabled;
     const isYoutubeUrl = typeof url === 'string' ? /youtube\.com|youtu\.be/i.test(url) : 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 = format === 'markdown'
+        ? parseMarkdownMode(program.opts().markdownMode ??
+            program.opts().markdown ??
+            'auto')
+        : 'off';
     const requestedFirecrawlMode = parseFirecrawlMode(program.opts().firecrawl);
     const modelArg = typeof program.opts().model === 'string' ? program.opts().model : null;
     const { config, path: configPath } = loadSummarizeConfig({ env });
@@ -540,6 +606,12 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
     const anthropicConfigured = typeof anthropicApiKey === 'string' && anthropicApiKey.length > 0;
     const openrouterConfigured = typeof openrouterApiKey === 'string' && openrouterApiKey.length > 0;
     const openrouterOptions = openRouterProviders ? { providers: openRouterProviders } : undefined;
+    if (markdownModeExplicitlySet && format !== '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;
@@ -642,7 +714,7 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
             return streamMode;
         return isRichTty(stdout) ? 'on' : 'off';
     })();
-    const streamingEnabled = effectiveStreamMode === 'on' && !json && !extractOnly;
+    const streamingEnabled = effectiveStreamMode === 'on' && !json && !extractMode;
     const effectiveRenderMode = (() => {
         if (renderMode !== 'auto')
             return renderMode;
@@ -661,8 +733,8 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
         stderr.write(`metrics apify requests=${report.services.apify.requests}\n`);
         stderr.write(`metrics total tok(i/o/t)=${promptTokens ?? 'unknown'}/${completionTokens ?? 'unknown'}/${totalTokens ?? 'unknown'}\n`);
     };
-    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 && !json;
     let clearProgressBeforeStdout = null;
@@ -697,11 +769,6 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
         if (!hasRequiredKey) {
             throw new Error(`Missing ${requiredKeyEnv} for model ${parsedModel.canonical}. Set the env var or choose a different --model.`);
         }
-        assertProviderSupportsAttachment({
-            provider: parsedModel.provider,
-            modelId: parsedModel.canonical,
-            attachment: { part: attachment.part, mediaType: attachment.mediaType },
-        });
         const modelResolution = await resolveModelIdForLlmCall({
             parsedModel,
             apiKeys: { googleApiKey: apiKeysForLlm.googleApiKey },
@@ -719,14 +786,114 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
         if (textContent && textContent.bytes > MAX_TEXT_BYTES_DEFAULT) {
             throw new Error(`Text file too large (${formatBytes(textContent.bytes)}). Limit is ${formatBytes(MAX_TEXT_BYTES_DEFAULT)}.`);
         }
+        const fileBytes = getFileBytesFromAttachment(attachment);
+        const canPreprocessWithMarkitdown = format === 'markdown' &&
+            preprocessMode !== 'off' &&
+            hasUvxCli(env) &&
+            attachment.part.type === 'file' &&
+            fileBytes !== null &&
+            shouldMarkitdownConvertMediaType(attachment.mediaType);
         const summaryLengthTarget = lengthArg.kind === 'preset' ? lengthArg.preset : { maxCharacters: lengthArg.maxCharacters };
-        const promptText = buildFileSummaryPrompt({
-            filename: attachment.filename,
-            mediaType: attachment.mediaType,
-            summaryLength: summaryLengthTarget,
-            contentLength: textContent?.content.length ?? null,
-        });
-        const promptPayload = buildAssetPromptPayload({ promptText, attachment, textContent });
+        let promptText = '';
+        const buildAttachmentPromptPayload = () => {
+            promptText = buildFileSummaryPrompt({
+                filename: attachment.filename,
+                mediaType: attachment.mediaType,
+                summaryLength: summaryLengthTarget,
+                contentLength: textContent?.content.length ?? null,
+            });
+            return buildAssetPromptPayload({ promptText, attachment, textContent });
+        };
+        const buildMarkitdownPromptPayload = (markdown) => {
+            promptText = buildFileTextSummaryPrompt({
+                filename: attachment.filename,
+                originalMediaType: attachment.mediaType,
+                contentMediaType: 'text/markdown',
+                summaryLength: summaryLengthTarget,
+                contentLength: markdown.length,
+            });
+            return `${promptText}\n\n---\n\n${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: env.UVX_PATH,
+                    timeoutMs,
+                    env,
+                    execFileImpl,
+                });
+            }
+            catch (error) {
+                const message = error instanceof Error ? error.message : String(error);
+                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 (error) {
+                if (!canPreprocessWithMarkitdown) {
+                    if (format === 'markdown' &&
+                        preprocessMode !== 'off' &&
+                        attachment.part.type === 'file' &&
+                        shouldMarkitdownConvertMediaType(attachment.mediaType) &&
+                        !hasUvxCli(env)) {
+                        throw withUvxTip(error, env);
+                    }
+                    throw error;
+                }
+                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: env.UVX_PATH,
+                        timeoutMs,
+                        env,
+                        execFileImpl,
+                    });
+                }
+                catch (markitdownError) {
+                    if (preprocessMode === 'auto') {
+                        throw error;
+                    }
+                    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) &&
@@ -1128,12 +1295,21 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
     if (!url) {
         throw new Error('Only HTTP and HTTPS URLs can be summarized');
     }
-    const firecrawlMode = requestedFirecrawlMode;
+    const wantsMarkdown = format === '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'
@@ -1150,16 +1326,16 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
                 : parsedModelForLlm.provider === 'anthropic'
                     ? 'ANTHROPIC_API_KEY'
                     : 'OPENAI_API_KEY';
-        throw new Error(`--markdown llm requires ${required} for model ${parsedModelForLlm.canonical}`);
+        throw new Error(`--markdown-mode llm requires ${required} for model ${parsedModelForLlm.canonical}`);
     }
-    writeVerbose(stderr, verbose, `config url=${url} timeoutMs=${timeoutMs} youtube=${youtubeMode} firecrawl=${firecrawlMode} length=${lengthArg.kind === 'preset' ? lengthArg.preset : `${lengthArg.maxCharacters} chars`} maxOutputTokens=${formatOptionalNumber(maxOutputTokensArg)} json=${json} extractOnly=${extractOnly} markdown=${effectiveMarkdownMode} model=${model} stream=${effectiveStreamMode} render=${effectiveRenderMode}`, verboseColor);
+    writeVerbose(stderr, verbose, `config url=${url} timeoutMs=${timeoutMs} youtube=${youtubeMode} firecrawl=${firecrawlMode} length=${lengthArg.kind === 'preset' ? lengthArg.preset : `${lengthArg.maxCharacters} chars`} maxOutputTokens=${formatOptionalNumber(maxOutputTokensArg)} json=${json} extract=${extractMode} format=${format} preprocess=${preprocessMode} markdownMode=${markdownMode} model=${model} stream=${effectiveStreamMode} render=${effectiveRenderMode}`, verboseColor);
     writeVerbose(stderr, verbose, `configFile path=${formatOptionalString(configPath)} model=${formatOptionalString(config?.model ?? null)}`, verboseColor);
     writeVerbose(stderr, verbose, `env xaiKey=${xaiConfigured} openaiKey=${Boolean(apiKey)} googleKey=${googleConfigured} anthropicKey=${anthropicConfigured} openrouterKey=${openrouterConfigured} apifyToken=${Boolean(apifyToken)} firecrawlKey=${firecrawlConfigured}`, verboseColor);
     writeVerbose(stderr, verbose, `markdown requested=${markdownRequested} provider=${markdownProvider}`, verboseColor);
     const scrapeWithFirecrawl = firecrawlConfigured && firecrawlMode !== 'off'
         ? createFirecrawlScraper({ apiKey: firecrawlApiKey, fetchImpl: trackedFetch })
         : null;
-    const convertHtmlToMarkdown = markdownRequested && (effectiveMarkdownMode === 'llm' || markdownProvider !== 'none')
+    const llmHtmlToMarkdown = markdownRequested && (effectiveMarkdownMode === 'llm' || markdownProvider !== 'none')
         ? createHtmlToMarkdownConverter({
             modelId: model,
             xaiApiKey: xaiConfigured ? xaiApiKey : null,
@@ -1174,6 +1350,46 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
             },
         })
         : null;
+    const markitdownHtmlToMarkdown = markdownRequested && preprocessMode !== 'off' && hasUvxCli(env)
+        ? 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: env.UVX_PATH,
+                timeoutMs: args.timeoutMs,
+                env,
+                execFileImpl,
+            });
+        }
+        : null;
+    const convertHtmlToMarkdown = markdownRequested
+        ? async (args) => {
+            if (effectiveMarkdownMode === 'llm') {
+                if (!llmHtmlToMarkdown) {
+                    throw new Error('No HTML→Markdown converter configured');
+                }
+                return llmHtmlToMarkdown(args);
+            }
+            if (llmHtmlToMarkdown) {
+                try {
+                    return await llmHtmlToMarkdown(args);
+                }
+                catch (error) {
+                    if (!markitdownHtmlToMarkdown)
+                        throw error;
+                    return await markitdownHtmlToMarkdown(args);
+                }
+            }
+            if (markitdownHtmlToMarkdown) {
+                return await markitdownHtmlToMarkdown(args);
+            }
+            throw new Error('No HTML→Markdown converter configured');
+        }
+        : null;
     const readTweetWithBirdClient = hasBirdCli(env)
         ? ({ url, timeoutMs }) => readTweetWithBird({ url, timeoutMs, env })
         : null;
@@ -1367,7 +1583,7 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
         const viaSourceLabel = viaSources.length > 0 ? `, ${viaSources.join('+')}` : '';
         if (progressEnabled) {
             websiteProgress?.stop?.();
-            spinner.setText(extractOnly
+            spinner.setText(extractMode
                 ? `Extracted (${extractedContentSize}${viaSourceLabel})`
                 : `Summarizing (sent ${extractedContentSize}${viaSourceLabel})…`);
         }
@@ -1378,6 +1594,14 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
         writeVerbose(stderr, verbose, `extract transcript textProvided=${extracted.diagnostics.transcript.textProvided} provider=${formatOptionalString(extracted.diagnostics.transcript.provider ?? null)} 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(env)) {
+            stderr.write(`${UVX_TIP}\n`);
+        }
         const isYouTube = extracted.siteName === 'YouTube';
         const prompt = buildLinkSummaryPrompt({
             url: extracted.url,
@@ -1391,7 +1615,7 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
             summaryLength: lengthArg.kind === 'preset' ? lengthArg.preset : { maxCharacters: lengthArg.maxCharacters },
             shares: [],
         });
-        if (extractOnly) {
+        if (extractMode) {
             clearProgressForStdout();
             if (json) {
                 const finishReport = shouldComputeReport ? await buildReport() : null;
@@ -1402,6 +1626,7 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
                         timeoutMs,
                         youtube: youtubeMode,
                         firecrawl: firecrawlMode,
+                        format,
                         markdown: effectiveMarkdownMode,
                         length: lengthArg.kind === 'preset'
                             ? { kind: 'preset', preset: lengthArg.preset }
@@ -1472,6 +1697,7 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
                         timeoutMs,
                         youtube: youtubeMode,
                         firecrawl: firecrawlMode,
+                        format,
                         markdown: effectiveMarkdownMode,
                         length: lengthArg.kind === 'preset'
                             ? { kind: 'preset', preset: lengthArg.preset }
@@ -1744,6 +1970,7 @@ export async function runCli(argv, { env, fetch, stdout, stderr }) {
                     timeoutMs,
                     youtube: youtubeMode,
                     firecrawl: firecrawlMode,
+                    format,
                     markdown: effectiveMarkdownMode,
                     length: lengthArg.kind === 'preset'
                         ? { kind: 'preset', preset: lengthArg.preset }