OpenClaw \u53ef\u4ee5\u5728\u56de\u590d\u6d41\u7a0b\u8fd0\u884c\u4e4b\u524d\u6458\u8981\u5165\u7ad9\u5a92\u4f53<\/strong>\uff08\u56fe\u7247\/\u97f3\u9891\/\u89c6\u9891\uff09\u3002\u5b83\u4f1a\u81ea\u52a8\u68c0\u6d4b\u672c\u5730\u5de5\u5177\u6216\u63d0\u4f9b\u5546\u5bc6\u94a5\u662f\u5426\u53ef\u7528\uff0c\u5e76\u4e14\u53ef\u4ee5\u7981\u7528\u6216\u81ea\u5b9a\u4e49\u3002\u5982\u679c\u7406\u89e3\u5173\u95ed\uff0c\u6a21\u578b\u4ecd\u7136\u4f1a\u50cf\u5f80\u5e38\u4e00\u6837\u63a5\u6536\u539f\u59cb\u6587\u4ef6\/URL\u3002<\/p>\n \u5982\u679c\u7406\u89e3\u5931\u8d25\u6216\u88ab\u7981\u7528\uff0c\u56de\u590d\u6d41\u7a0b\u7ee7\u7eed<\/strong>\u4f7f\u7528\u539f\u59cb\u6b63\u6587 + \u9644\u4ef6\u3002<\/p>\n “`json5 theme={“theme”:{“light”:”min-light”,”dark”:”min-dark”}} “`json5 theme={“theme”:{“light”:”min-light”,”dark”:”min-dark”}} \u6ce8\u610f\uff1a\u4e8c\u8fdb\u5236\u6587\u4ef6\u68c0\u6d4b\u5728 macOS\/Linux\/Windows \u4e0a\u662f\u5c3d\u529b\u800c\u4e3a\u7684\uff1b\u786e\u4fdd CLI \u5728 \u5982\u679c\u4f60\u8bbe\u7f6e\u4e86 \u5bf9\u4e8e CLI \u6761\u76ee\uff0c\u663e\u5f0f\u8bbe\u7f6e \u76ee\u6807<\/h2>\n
\n
\u9ad8\u5c42\u884c\u4e3a<\/h2>\n
\n
MediaPaths<\/code>\u3001MediaUrls<\/code>\u3001MediaTypes<\/code>\uff09\u3002<\/li>\n
\n * Body<\/code> \u53d8\u4e3a [Image]<\/code>\u3001[Audio]<\/code> \u6216 [Video]<\/code> \u5757\u3002
\n * \u97f3\u9891\u8bbe\u7f6e {{Transcript}}<\/code>\uff1b\u547d\u4ee4\u89e3\u6790\u5728\u6709\u6807\u9898\u6587\u672c\u65f6\u4f7f\u7528\u6807\u9898\u6587\u672c\uff0c\u5426\u5219\u4f7f\u7528\u8f6c\u5f55\u3002
\n * \u6807\u9898\u4f5c\u4e3a User text:<\/code> \u4fdd\u7559\u5728\u5757\u5185\u3002<\/li>\n<\/ol>\n\u914d\u7f6e\u6982\u8ff0<\/h2>\n
tools.media<\/code> \u652f\u6301\u5171\u4eab\u6a21\u578b<\/strong>\u52a0\u4e0a\u6bcf\u80fd\u529b\u8986\u76d6\uff1a<\/p>\n\n
tools.media.models<\/code>\uff1a\u5171\u4eab\u6a21\u578b\u5217\u8868\uff08\u4f7f\u7528 capabilities<\/code> \u6765\u9650\u5b9a\uff09\u3002<\/li>\ntools.media.image<\/code> \/ tools.media.audio<\/code> \/ tools.media.video<\/code>\uff1a<\/li>\nprompt<\/code>\u3001maxChars<\/code>\u3001maxBytes<\/code>\u3001timeoutSeconds<\/code>\u3001language<\/code>\uff09<\/li>\nbaseUrl<\/code>\u3001headers<\/code>\u3001providerOptions<\/code>\uff09<\/li>\ntools.media.audio.providerOptions.deepgram<\/code> \u914d\u7f6e Deepgram \u97f3\u9891\u9009\u9879<\/li>\nmodels<\/code> \u5217\u8868<\/strong>\uff08\u4f18\u5148\u4e8e\u5171\u4eab\u6a21\u578b\uff09<\/li>\nattachments<\/code> \u7b56\u7565\uff08mode<\/code>\u3001maxAttachments<\/code>\u3001prefer<\/code>\uff09<\/li>\nscope<\/code>\uff08\u53ef\u9009\u7684\u6309\u6e20\u9053\/\u804a\u5929\u7c7b\u578b\/\u4f1a\u8bdd\u952e\u9650\u5b9a\uff09<\/li>\ntools.media.concurrency<\/code>\uff1a\u6700\u5927\u5e76\u53d1\u80fd\u529b\u8fd0\u884c\u6570\uff08\u9ed8\u8ba4 2<\/strong>\uff09\u3002<\/li>\n<\/ul>\n
\n{
\n tools: {
\n media: {
\n models: [
\n \/ \u5171\u4eab\u5217\u8868 <\/em>\/
\n ],
\n image: {
\n \/ \u53ef\u9009\u8986\u76d6 <\/em>\/
\n },
\n audio: {
\n \/ \u53ef\u9009\u8986\u76d6 <\/em>\/
\n },
\n video: {
\n \/ \u53ef\u9009\u8986\u76d6 <\/em>\/
\n },
\n },
\n },
\n}<\/p>\n\n### \u6a21\u578b\u6761\u76ee\n\n\u6bcf\u4e2a `models[]` \u6761\u76ee\u53ef\u4ee5\u662f**\u63d0\u4f9b\u5546**\u6216 **CLI**\uff1a\n\n```json5 theme={"theme":{"light":"min-light","dark":"min-dark"}}\n{\n type: "provider", \/\/ \u7701\u7565\u65f6\u9ed8\u8ba4\n provider: "openai",\n model: "gpt-5.2",\n prompt: "Describe the image in <= 500 chars.",\n maxChars: 500,\n maxBytes: 10485760,\n timeoutSeconds: 60,\n capabilities: ["image"], \/\/ \u53ef\u9009\uff0c\u7528\u4e8e\u591a\u6a21\u6001\u6761\u76ee\n profile: "vision-profile",\n preferredProfile: "vision-fallback",\n}\n<\/code><\/pre>\n
\n{
\n type: “cli”,
\n command: “gemini”,
\n args: [
\n “-m”,
\n “gemini-3-flash”,
\n “–allowed-tools”,
\n “read_file”,
\n “Read the media at {{MediaPath}} and describe it in <= {{MaxChars}} characters.”,
\n ],
\n maxChars: 500,
\n maxBytes: 52428800,
\n timeoutSeconds: 120,
\n capabilities: [“video”, “image”],
\n}<\/p>\n\nCLI \u6a21\u677f\u8fd8\u53ef\u4ee5\u4f7f\u7528\uff1a\n\n* `{{MediaDir}}`\uff08\u5305\u542b\u5a92\u4f53\u6587\u4ef6\u7684\u76ee\u5f55\uff09\n* `{{OutputDir}}`\uff08\u4e3a\u672c\u6b21\u8fd0\u884c\u521b\u5efa\u7684\u4e34\u65f6\u76ee\u5f55\uff09\n* `{{OutputBase}}`\uff08\u4e34\u65f6\u6587\u4ef6\u57fa\u7840\u8def\u5f84\uff0c\u65e0\u6269\u5c55\u540d\uff09\n\n## \u9ed8\u8ba4\u503c\u548c\u9650\u5236\n\n\u63a8\u8350\u9ed8\u8ba4\u503c\uff1a\n\n* `maxChars`\uff1a\u56fe\u7247\/\u89c6\u9891\u4e3a **500**\uff08\u7b80\u77ed\uff0c\u9002\u5408\u547d\u4ee4\uff09\n* `maxChars`\uff1a\u97f3\u9891**\u4e0d\u8bbe\u7f6e**\uff08\u5b8c\u6574\u8f6c\u5f55\uff0c\u9664\u975e\u4f60\u8bbe\u7f6e\u9650\u5236\uff09\n* `maxBytes`\uff1a\n * \u56fe\u7247\uff1a**10MB**\n * \u97f3\u9891\uff1a**20MB**\n * \u89c6\u9891\uff1a**50MB**\n\n\u89c4\u5219\uff1a\n\n* \u5982\u679c\u5a92\u4f53\u8d85\u8fc7 `maxBytes`\uff0c\u8be5\u6a21\u578b\u88ab\u8df3\u8fc7\uff0c**\u5c1d\u8bd5\u4e0b\u4e00\u4e2a\u6a21\u578b**\u3002\n* \u5982\u679c\u6a21\u578b\u8fd4\u56de\u8d85\u8fc7 `maxChars`\uff0c\u8f93\u51fa\u88ab\u622a\u65ad\u3002\n* `prompt` \u9ed8\u8ba4\u4e3a\u7b80\u5355\u7684"Describe the {media}."\u52a0\u4e0a `maxChars` \u6307\u5bfc\uff08\u4ec5\u56fe\u7247\/\u89c6\u9891\uff09\u3002\n* \u5982\u679c `<capability>.enabled: true` \u4f46\u672a\u914d\u7f6e\u6a21\u578b\uff0c\u5f53\u63d0\u4f9b\u5546\u652f\u6301\u8be5\u80fd\u529b\u65f6\uff0cOpenClaw \u5c1d\u8bd5**\u6d3b\u52a8\u7684\u56de\u590d\u6a21\u578b**\u3002\n\n### \u81ea\u52a8\u68c0\u6d4b\u5a92\u4f53\u7406\u89e3\uff08\u9ed8\u8ba4\uff09\n\n\u5982\u679c `tools.media.<capability>.enabled` **\u672a**\u8bbe\u7f6e\u4e3a `false` \u4e14\u4f60\u6ca1\u6709\u914d\u7f6e\u6a21\u578b\uff0cOpenClaw \u6309\u4ee5\u4e0b\u987a\u5e8f\u81ea\u52a8\u68c0\u6d4b\u5e76**\u5728\u7b2c\u4e00\u4e2a\u53ef\u7528\u9009\u9879\u5904\u505c\u6b62**\uff1a\n\n1. **\u672c\u5730 CLI**\uff08\u4ec5\u97f3\u9891\uff1b\u5982\u679c\u5df2\u5b89\u88c5\uff09\n * `sherpa-onnx-offline`\uff08\u9700\u8981\u5e26\u6709 encoder\/decoder\/joiner\/tokens \u7684 `SHERPA_ONNX_MODEL_DIR`\uff09\n * `whisper-cli`\uff08`whisper-cpp`\uff1b\u4f7f\u7528 `WHISPER_CPP_MODEL` \u6216\u6346\u7ed1\u7684 tiny \u6a21\u578b\uff09\n * `whisper`\uff08Python CLI\uff1b\u81ea\u52a8\u4e0b\u8f7d\u6a21\u578b\uff09\n2. **Gemini CLI**\uff08`gemini`\uff09\u4f7f\u7528 `read_many_files`\n3. **\u63d0\u4f9b\u5546\u5bc6\u94a5**\n * \u97f3\u9891\uff1aOpenAI \u2192 Groq \u2192 Deepgram \u2192 Google\n * \u56fe\u7247\uff1aOpenAI \u2192 Anthropic \u2192 Google \u2192 MiniMax\n * \u89c6\u9891\uff1aGoogle\n\n\u8981\u7981\u7528\u81ea\u52a8\u68c0\u6d4b\uff0c\u8bbe\u7f6e\uff1a\n\n```json5 theme={"theme":{"light":"min-light","dark":"min-dark"}}\n{\n tools: {\n media: {\n audio: {\n enabled: false,\n },\n },\n },\n}\n<\/code><\/pre>\nPATH<\/code> \u4e0a\uff08\u6211\u4eec\u4f1a\u5c55\u5f00 ~<\/code>\uff09\uff0c\u6216\u8bbe\u7f6e\u5e26\u6709\u5b8c\u6574\u547d\u4ee4\u8def\u5f84\u7684\u663e\u5f0f CLI \u6a21\u578b\u3002<\/p>\n\u80fd\u529b\uff08\u53ef\u9009\uff09<\/h2>\n
capabilities<\/code>\uff0c\u8be5\u6761\u76ee\u4ec5\u5bf9\u8fd9\u4e9b\u5a92\u4f53\u7c7b\u578b\u8fd0\u884c\u3002\u5bf9\u4e8e\u5171\u4eab\u5217\u8868\uff0cOpenClaw \u53ef\u4ee5\u63a8\u65ad\u9ed8\u8ba4\u503c\uff1a<\/p>\n\n
openai<\/code>\u3001anthropic<\/code>\u3001minimax<\/code>\uff1a\u56fe\u7247<\/strong><\/li>\ngoogle<\/code>\uff08Gemini API\uff09\uff1a\u56fe\u7247 + \u97f3\u9891 + \u89c6\u9891<\/strong><\/li>\ngroq<\/code>\uff1a\u97f3\u9891<\/strong><\/li>\ndeepgram<\/code>\uff1a\u97f3\u9891<\/strong><\/li>\n<\/ul>\ncapabilities<\/code><\/strong> \u4ee5\u907f\u514d\u610f\u5916\u5339\u914d\u3002\u5982\u679c\u4f60\u7701\u7565 capabilities<\/code>\uff0c\u8be5\u6761\u76ee\u5bf9\u5b83\u51fa\u73b0\u7684\u5217\u8868\u90fd\u7b26\u5408\u6761\u4ef6\u3002<\/p>\n\u63d0\u4f9b\u5546\u652f\u6301\u77e9\u9635\uff08OpenClaw \u96c6\u6210\uff09<\/h2>\n