@mindstudio-ai/remy 0.1.178 → 0.1.179

package/dist/headless.js

@@ -404,7 +404,7 @@ async function* streamChat(params) {
   const url = `${baseUrl2}/_internal/v2/agent/remy/chat`;
   const startTime = Date.now();
   const subAgentId = rest.subAgentId;
-  const requestBody = { ...rest, ...model && { modelId: model } };
+  const requestBody = { ...rest, modelId: model };
   log2.info("API request", {
     requestId,
     ...subAgentId && { subAgentId },
@@ -2787,13 +2787,8 @@ var queryDatabaseTool = {
 };
 
 // src/subagents/common/analyzeImage.ts
-var VISION_MODEL = "claude-4-6-sonnet";
-var VISION_MODEL_OVERRIDE = {
-  model: VISION_MODEL,
-  config: { thinkingBudget: "off" }
-};
 async function analyzeImage(params) {
-  const { prompt, imageUrl, timeout = 2e5, onLog } = params;
+  const { prompt, imageUrl, model, timeout = 2e5, onLog } = params;
   return runMindstudioCli(
     [
       "analyze-image",
@@ -2802,7 +2797,7 @@ async function analyzeImage(params) {
       "--image-url",
       imageUrl,
       "--vision-model-override",
-      JSON.stringify(VISION_MODEL_OVERRIDE)
+      JSON.stringify({ model })
     ],
     { outputKey: "analysis", timeout, onLog }
   );
@@ -2830,13 +2825,14 @@ ${TEXT_WRAP_DISCLAIMER}`;
   return p;
 }
 async function streamScreenshotAnalysis(opts) {
-  const { url, prompt, styleMap, onLog } = opts;
+  const { url, prompt, styleMap, onLog, model } = opts;
   onLog?.(JSON.stringify({ url, analysis: null }));
   const analysisPrompt = buildScreenshotAnalysisPrompt({ prompt, styleMap });
   let accumulated = "";
   const analysis = await analyzeImage({
     prompt: analysisPrompt,
     imageUrl: url,
+    model,
     onLog: (chunk) => {
       accumulated += chunk;
       onLog?.(JSON.stringify({ url, analysis: accumulated }));
@@ -2848,12 +2844,14 @@ async function captureAndAnalyzeScreenshot(promptOrOptions) {
   let prompt;
   let existingUrl;
   let onLog;
+  let model;
   let path12;
   if (typeof promptOrOptions === "object" && promptOrOptions !== null) {
     prompt = promptOrOptions.prompt;
     existingUrl = promptOrOptions.imageUrl;
     path12 = promptOrOptions.path;
     onLog = promptOrOptions.onLog;
+    model = promptOrOptions.model;
   } else {
     prompt = promptOrOptions;
   }
@@ -2878,11 +2876,17 @@ async function captureAndAnalyzeScreenshot(promptOrOptions) {
   if (prompt === false) {
     return url;
   }
+  if (!model) {
+    throw new Error(
+      "captureAndAnalyzeScreenshot: `model` is required when analysis is enabled"
+    );
+  }
   return streamScreenshotAnalysis({
     url,
     prompt: prompt || void 0,
     styleMap,
-    onLog
+    onLog,
+    model
   });
 }
 
@@ -3715,6 +3719,98 @@ ${appSpec}
   }
 }
 
+// src/models/surfaces.ts
+var MODEL_SURFACES = {
+  parent: {
+    default: "claude-4-7-opus",
+    label: "Remy",
+    description: "The main Remy agent you chat with about your product. Writes code and manages delegation to other agents.",
+    modelType: "text",
+    userPickable: true
+  },
+  visualDesignExpert: {
+    default: "claude-4-7-opus",
+    label: "Design Agent",
+    description: "Designs your product's interfaces, including components, layouts, typography, color, and visual identity.",
+    modelType: "text",
+    userPickable: true
+  },
+  productVision: {
+    default: "claude-4-6-sonnet",
+    label: "Roadmap Agent",
+    description: "Owns your product's roadmap and pitch deck. Helps decide what to build next and how to frame the big picture.",
+    modelType: "text",
+    userPickable: true
+  },
+  browserAutomation: {
+    default: "claude-4-6-sonnet",
+    label: "QA Agent",
+    description: "Tests features and UI flows in an automated browser to verify they work end to end.",
+    modelType: "text",
+    userPickable: true
+  },
+  codeSanityCheck: {
+    default: "claude-4-6-sonnet",
+    label: "Architecture Agent",
+    description: "Reviews the architecture and structure of code changes to avoid technical debt.",
+    modelType: "text",
+    userPickable: true
+  },
+  imageGeneration: {
+    default: "seedream-4.5",
+    label: "Image Generation",
+    description: "Creates images for your product \u2014 icons, illustrations, photos, and any other visual assets.",
+    modelType: "image_generation",
+    userPickable: true
+  },
+  imageAnalysis: {
+    default: "claude-4-6-sonnet",
+    label: "Image Analysis",
+    description: "Reads screenshots taken by the QA agent during automated browser tests. Other agents use their own built-in image analysis when they need to read images.",
+    modelType: "vision",
+    userPickable: true
+  },
+  conversationSummarizer: {
+    default: "claude-4-6-sonnet",
+    label: "Compaction Utility",
+    description: "Compresses long conversations into summaries to keep things responsive.",
+    modelType: "text",
+    userPickable: true
+  },
+  brandExtractor: {
+    default: "claude-4-6-sonnet",
+    label: "Brand Utility",
+    description: "Extracts your product's name, colors, and fonts from your spec for use in branded documents.",
+    modelType: "text",
+    userPickable: true
+  },
+  // Internal surface — not user-pickable. Remy uses this to rewrite design
+  // briefs into model-optimized image prompts before image generation.
+  imagePromptEnhancer: {
+    default: "claude-4-6-sonnet",
+    label: "Image Prompt Enhancer",
+    description: "Rewrites image briefs into model-optimized prompts before image generation.",
+    modelType: "text",
+    userPickable: false
+  }
+};
+var ALLOWED_MODELS_BY_TYPE = {
+  text: [
+    "claude-4-7-opus",
+    "claude-4-6-opus",
+    "claude-4-6-sonnet",
+    "gpt-5.5",
+    "gemini-3-pro",
+    "gemini-3.1-pro",
+    "gemini-3-flash"
+  ]
+  // vision: undefined — unconstrained
+  // image_generation: undefined — unconstrained
+};
+function resolveModel(surfaceId, models, fallback) {
+  return models?.[surfaceId] ?? fallback ?? MODEL_SURFACES[surfaceId].default;
+}
+
 // src/subagents/browserAutomation/index.ts
 var log6 = createLogger("browser-automation");
 var browserAutomationTool = {
@@ -3764,7 +3860,12 @@ var browserAutomationTool = {
             try {
               return await captureAndAnalyzeScreenshot({
                 path: _input.path,
-                onLog
+                onLog,
+                model: resolveModel(
+                  "imageAnalysis",
+                  context.models,
+                  context.model
+                )
               });
             } catch (err) {
               return `Error taking screenshot: ${err.message}`;
@@ -3773,7 +3874,7 @@ var browserAutomationTool = {
           return `Error: unknown local tool "${name}"`;
         },
         apiConfig: context.apiConfig,
-        model: context.models?.browserAutomation ?? context.model,
+        model: resolveModel("browserAutomation", context.models, context.model),
         subAgentId: "browserAutomation",
         signal: context.signal,
         parentToolId: context.toolCallId,
@@ -3791,6 +3892,13 @@ var browserAutomationTool = {
                 (s) => s.command === "screenshotViewport" && s.result?.url
               );
               if (screenshotSteps.length > 0) {
+                const visionOverride = {
+                  model: resolveModel(
+                    "imageAnalysis",
+                    context.models,
+                    context.model
+                  )
+                };
                 const batchInput = screenshotSteps.map((s) => ({
                   stepType: "analyzeImage",
                   step: {
@@ -3798,7 +3906,7 @@ var browserAutomationTool = {
                     prompt: buildScreenshotAnalysisPrompt({
                       styleMap: s.result.styleMap
                     }),
-                    visionModelOverride: VISION_MODEL_OVERRIDE
+                    visionModelOverride: visionOverride
                   }
                 }));
                 const batchResult = await runMindstudioCli(
@@ -3877,7 +3985,8 @@ var screenshotTool = {
         return await captureAndAnalyzeScreenshot({
           prompt: input.prompt,
           imageUrl: input.imageUrl,
-          onLog: context?.onLog
+          onLog: context?.onLog,
+          model: resolveModel("imageAnalysis", context?.models, context?.model)
         });
       }
       if (input.instructions && context) {
@@ -3899,7 +4008,8 @@ var screenshotTool = {
           url,
           prompt: input.prompt,
           styleMap,
-          onLog: context?.onLog
+          onLog: context?.onLog,
+          model: resolveModel("imageAnalysis", context?.models, context?.model)
         });
       }
       const release = await acquireBrowserLock();
@@ -3907,7 +4017,8 @@ var screenshotTool = {
         return await captureAndAnalyzeScreenshot({
           prompt: input.prompt,
           path: input.path,
-          onLog: context?.onLog
+          onLog: context?.onLog,
+          model: resolveModel("imageAnalysis", context?.models, context?.model)
         });
       } finally {
         release();
@@ -4118,7 +4229,7 @@ var definition3 = {
     required: ["url"]
   }
 };
-async function execute3(input, onLog) {
+async function execute3(input, onLog, context) {
   const url = input.url;
   const analysisPrompt = input.prompt || DESIGN_REFERENCE_PROMPT;
   const isImageUrl = /\.(png|jpe?g|webp|gif|svg|avif)(\?|$)/i.test(url);
@@ -4151,7 +4262,8 @@ async function execute3(input, onLog) {
   const analysis = await analyzeImage({
     prompt: analysisPrompt,
     imageUrl,
-    onLog
+    onLog,
+    model: resolveModel("imageAnalysis", context?.models, context?.model)
   });
   return JSON.stringify({ url: imageUrl, analysis });
 }
@@ -4181,7 +4293,7 @@ var definition4 = {
     required: ["imageUrl"]
   }
 };
-async function execute4(input, onLog) {
+async function execute4(input, onLog, context) {
   const imageUrl = input.imageUrl;
   const prompt = buildScreenshotAnalysisPrompt({
     prompt: input.prompt
@@ -4189,7 +4301,8 @@ async function execute4(input, onLog) {
   const analysis = await analyzeImage({
     prompt,
     imageUrl,
-    onLog
+    onLog,
+    model: resolveModel("imageAnalysis", context?.models, context?.model)
   });
   return JSON.stringify({ url: imageUrl, analysis });
 }
@@ -4243,7 +4356,8 @@ async function execute5(input, onLog, context) {
         url,
         prompt: input.prompt,
         styleMap,
-        onLog
+        onLog,
+        model: resolveModel("imageAnalysis", context?.models, context?.model)
       });
     } catch (err) {
       return `Error taking interactive screenshot: ${err.message}`;
@@ -4254,7 +4368,8 @@ async function execute5(input, onLog, context) {
     return await captureAndAnalyzeScreenshot({
       prompt: input.prompt,
       path: input.path,
-      onLog
+      onLog,
+      model: resolveModel("imageAnalysis", context?.models, context?.model)
     });
   } catch (err) {
     return `Error taking screenshot: ${err.message}`;
@@ -4271,16 +4386,11 @@ __export(generateImages_exports, {
 });
 
 // src/subagents/designExpert/tools/images/enhancePrompt.ts
-var ENHANCE_MODEL = "claude-4-6-sonnet";
-var MODEL_OVERRIDE = {
-  model: ENHANCE_MODEL,
-  config: { reasoning: "false" }
-};
 var SYSTEM_PROMPT = readAsset(
   "subagents/designExpert/tools/images/enhance-image-prompt.md"
 );
 async function enhanceImagePrompt(params) {
-  const { brief, width, height, transparentBackground, onLog } = params;
+  const { brief, width, height, transparentBackground, onLog, model } = params;
   const contextParts = [
     `Dimensions: ${width}x${height}${width > height ? " (landscape)" : width < height ? " (portrait)" : " (square)"}`
   ];
@@ -4305,7 +4415,7 @@ ${brief}
       "--message",
       message,
       "--model-override",
-      JSON.stringify(MODEL_OVERRIDE)
+      JSON.stringify({ model, config: { reasoning: "false" } })
     ],
     { outputKey: "content", timeout: 6e4, onLog, caller: "designExpert" }
   );
@@ -4315,7 +4425,15 @@ ${brief}
 // src/subagents/designExpert/tools/images/imageGenerator.ts
 var ANALYZE_PROMPT = 'You are reviewing this image for a visual designer sourcing assets for a project. Describe: what the image depicts, the mood and color palette, how the lighting and composition work, any text present in the image, whether there are any issues (artifacts, distortions), and how it could be used in a layout for an app or website. Be concise and practical. Respond only with your analysis as Markdown (starting with the title "Asset Review") and absolutely no other text. Do not use emojis - use unicode if you need symbols.';
 async function generateImageAssets(opts) {
-  const { prompts, sourceImages, transparentBackground, onLog } = opts;
+  const {
+    prompts,
+    sourceImages,
+    transparentBackground,
+    onLog,
+    imageGenerationModel: genModel,
+    imageAnalysisModel,
+    imagePromptEnhancerModel
+  } = opts;
   const width = opts.width || 2048;
   const height = opts.height || 2048;
   const config = { width, height };
@@ -4330,7 +4448,8 @@ async function generateImageAssets(opts) {
         width,
         height,
         transparentBackground,
-        onLog
+        onLog,
+        model: imagePromptEnhancerModel
       })
     )
   );
@@ -4339,7 +4458,7 @@ async function generateImageAssets(opts) {
     const step = JSON.stringify({
       prompt: enhancedPrompts[0],
       imageModelOverride: {
-        model: "seedream-4.5",
+        model: genModel,
         config
       }
     });
@@ -4358,7 +4477,7 @@ async function generateImageAssets(opts) {
       step: {
         prompt,
         imageModelOverride: {
-          model: "seedream-4.5",
+          model: genModel,
           config
         }
       }
@@ -4410,7 +4529,8 @@ async function generateImageAssets(opts) {
       const analysis = await analyzeImage({
         prompt: ANALYZE_PROMPT,
         imageUrl: url,
-        onLog
+        onLog,
+        model: imageAnalysisModel
       });
       return {
         url,
@@ -4456,13 +4576,28 @@ var definition6 = {
     required: ["prompts"]
   }
 };
-async function execute6(input, onLog) {
+async function execute6(input, onLog, context) {
   return generateImageAssets({
     prompts: input.prompts,
     width: input.width,
     height: input.height,
     transparentBackground: input.transparentBackground,
-    onLog
+    onLog,
+    imageGenerationModel: resolveModel(
+      "imageGeneration",
+      context?.models,
+      context?.model
+    ),
+    imageAnalysisModel: resolveModel(
+      "imageAnalysis",
+      context?.models,
+      context?.model
+    ),
+    imagePromptEnhancerModel: resolveModel(
+      "imagePromptEnhancer",
+      context?.models,
+      context?.model
+    )
   });
 }
 
@@ -4509,14 +4644,29 @@ var definition7 = {
     required: ["prompts", "sourceImages"]
   }
 };
-async function execute7(input, onLog) {
+async function execute7(input, onLog, context) {
   return generateImageAssets({
     prompts: input.prompts,
     sourceImages: input.sourceImages,
     width: input.width,
     height: input.height,
     transparentBackground: input.transparentBackground,
-    onLog
+    onLog,
+    imageGenerationModel: resolveModel(
+      "imageGeneration",
+      context?.models,
+      context?.model
+    ),
+    imageAnalysisModel: resolveModel(
+      "imageAnalysis",
+      context?.models,
+      context?.model
+    ),
+    imagePromptEnhancerModel: resolveModel(
+      "imagePromptEnhancer",
+      context?.models,
+      context?.model
+    )
   });
 }
 
@@ -4989,7 +5139,7 @@ var designExpertTool = {
         );
       },
       apiConfig: context.apiConfig,
-      model: context.models?.visualDesignExpert ?? context.model,
+      model: resolveModel("visualDesignExpert", context.models, context.model),
       subAgentId: "visualDesignExpert",
       signal: context.signal,
       parentToolId: context.toolCallId,
@@ -5207,7 +5357,7 @@ var productVisionTool = {
         return executeVisionTool(name, input2, childCtx);
       },
       apiConfig: context.apiConfig,
-      model: context.models?.productVision ?? context.model,
+      model: resolveModel("productVision", context.models, context.model),
       subAgentId: "productVision",
       signal: context.signal,
       parentToolId: context.toolCallId,
@@ -5315,7 +5465,7 @@ var codeSanityCheckTool = {
       externalTools: /* @__PURE__ */ new Set(),
       executeTool: (name, toolInput) => executeTool(name, toolInput, context),
       apiConfig: context.apiConfig,
-      model: context.models?.codeSanityCheck ?? context.model,
+      model: resolveModel("codeSanityCheck", context.models, context.model),
       subAgentId: "codeSanityCheck",
       signal: context.signal,
       parentToolId: context.toolCallId,
@@ -5455,7 +5605,7 @@ function triggerCompaction(state, apiConfig, opts = {}) {
     apiConfig,
     system,
     tools2,
-    state.models?.conversationSummarizer ?? model
+    resolveModel("conversationSummarizer", state.models, model)
   ).then((summaries) => {
     pendingSummaries.push(...summaries);
     listener?.({ type: "complete", requestId });
@@ -6252,7 +6402,7 @@ async function runTurn(params) {
         onEvent({ type: "tool_input_delta", id, name, result: content });
       }
     }
-    const parentModel = state.models?.parent ?? model;
+    const parentModel = resolveModel("parent", state.models, model);
     try {
       for await (const event of streamChatWithRetry(
         {
@@ -6596,7 +6746,10 @@ async function runTurn(params) {
           isError: r.isError
         });
         if (!r.isError && BRAND_TRIGGERING_TOOLS.has(tc.name)) {
-          triggerBrandExtraction(apiConfig);
+          triggerBrandExtraction(
+            apiConfig,
+            resolveModel("brandExtractor", state.models, model)
+          );
         }
         return r;
       })
@@ -7039,12 +7192,14 @@ var HeadlessSession = class {
       this.emit("session_restored", {
         messageCount: this.state.messages.length,
         ...this.state.models && { models: this.state.models },
+        modelSurfaces: MODEL_SURFACES,
+        allowedModelsByType: ALLOWED_MODELS_BY_TYPE,
         ...this.queueFields()
       });
     }
     triggerBrandExtraction(
       this.config,
-      this.state.models?.brandExtractor ?? this.opts.model
+      resolveModel("brandExtractor", this.state.models, this.opts.model)
     );
     this.toolRegistry.onEvent = this.onEvent;
     setCompactionListener((event) => {
@@ -7722,6 +7877,8 @@ var HeadlessSession = class {
         running: this.running,
         ...this.running && this.currentRequestId ? { currentRequestId: this.currentRequestId } : {},
         ...this.state.models && { models: this.state.models },
+        modelSurfaces: MODEL_SURFACES,
+        allowedModelsByType: ALLOWED_MODELS_BY_TYPE,
         ...this.queueFields()
       }));
       return;

package/dist/index.js

@@ -90,7 +90,7 @@ async function* streamChat(params) {
   const url = `${baseUrl2}/_internal/v2/agent/remy/chat`;
   const startTime = Date.now();
   const subAgentId = rest.subAgentId;
-  const requestBody = { ...rest, ...model && { modelId: model } };
+  const requestBody = { ...rest, modelId: model };
   log.info("API request", {
     requestId,
     ...subAgentId && { subAgentId },
@@ -2024,6 +2024,104 @@ var init_prompt = __esm({
   }
 });
 
+// src/models/surfaces.ts
+function resolveModel(surfaceId, models, fallback) {
+  return models?.[surfaceId] ?? fallback ?? MODEL_SURFACES[surfaceId].default;
+}
+var MODEL_SURFACES, ALLOWED_MODELS_BY_TYPE;
+var init_surfaces = __esm({
+  "src/models/surfaces.ts"() {
+    "use strict";
+    MODEL_SURFACES = {
+      parent: {
+        default: "claude-4-7-opus",
+        label: "Remy",
+        description: "The main Remy agent you chat with about your product. Writes code and manages delegation to other agents.",
+        modelType: "text",
+        userPickable: true
+      },
+      visualDesignExpert: {
+        default: "claude-4-7-opus",
+        label: "Design Agent",
+        description: "Designs your product's interfaces, including components, layouts, typography, color, and visual identity.",
+        modelType: "text",
+        userPickable: true
+      },
+      productVision: {
+        default: "claude-4-6-sonnet",
+        label: "Roadmap Agent",
+        description: "Owns your product's roadmap and pitch deck. Helps decide what to build next and how to frame the big picture.",
+        modelType: "text",
+        userPickable: true
+      },
+      browserAutomation: {
+        default: "claude-4-6-sonnet",
+        label: "QA Agent",
+        description: "Tests features and UI flows in an automated browser to verify they work end to end.",
+        modelType: "text",
+        userPickable: true
+      },
+      codeSanityCheck: {
+        default: "claude-4-6-sonnet",
+        label: "Architecture Agent",
+        description: "Reviews the architecture and structure of code changes to avoid technical debt.",
+        modelType: "text",
+        userPickable: true
+      },
+      imageGeneration: {
+        default: "seedream-4.5",
+        label: "Image Generation",
+        description: "Creates images for your product \u2014 icons, illustrations, photos, and any other visual assets.",
+        modelType: "image_generation",
+        userPickable: true
+      },
+      imageAnalysis: {
+        default: "claude-4-6-sonnet",
+        label: "Image Analysis",
+        description: "Reads screenshots taken by the QA agent during automated browser tests. Other agents use their own built-in image analysis when they need to read images.",
+        modelType: "vision",
+        userPickable: true
+      },
+      conversationSummarizer: {
+        default: "claude-4-6-sonnet",
+        label: "Compaction Utility",
+        description: "Compresses long conversations into summaries to keep things responsive.",
+        modelType: "text",
+        userPickable: true
+      },
+      brandExtractor: {
+        default: "claude-4-6-sonnet",
+        label: "Brand Utility",
+        description: "Extracts your product's name, colors, and fonts from your spec for use in branded documents.",
+        modelType: "text",
+        userPickable: true
+      },
+      // Internal surface — not user-pickable. Remy uses this to rewrite design
+      // briefs into model-optimized image prompts before image generation.
+      imagePromptEnhancer: {
+        default: "claude-4-6-sonnet",
+        label: "Image Prompt Enhancer",
+        description: "Rewrites image briefs into model-optimized prompts before image generation.",
+        modelType: "text",
+        userPickable: false
+      }
+    };
+    ALLOWED_MODELS_BY_TYPE = {
+      text: [
+        "claude-4-7-opus",
+        "claude-4-6-opus",
+        "claude-4-6-sonnet",
+        "gpt-5.5",
+        "gemini-3-pro",
+        "gemini-3.1-pro",
+        "gemini-3-flash"
+      ]
+      // vision: undefined — unconstrained
+      // image_generation: undefined — unconstrained
+    };
+  }
+});
+
 // src/compaction/trigger.ts
 function getPendingSummaries() {
   return pendingSummaries.splice(0);
@@ -2044,7 +2142,7 @@ function triggerCompaction(state, apiConfig, opts = {}) {
     apiConfig,
     system,
     tools2,
-    state.models?.conversationSummarizer ?? model
+    resolveModel("conversationSummarizer", state.models, model)
   ).then((summaries) => {
     pendingSummaries.push(...summaries);
     listener?.({ type: "complete", requestId });
@@ -2067,6 +2165,7 @@ var init_trigger = __esm({
     init_prompt();
     init_tools6();
     init_logger();
+    init_surfaces();
     log3 = createLogger("compaction:trigger");
     pendingSummaries = [];
     inflightCompaction = null;
@@ -3085,7 +3184,7 @@ var init_queryDatabase = __esm({
 
 // src/subagents/common/analyzeImage.ts
 async function analyzeImage(params) {
-  const { prompt, imageUrl, timeout = 2e5, onLog } = params;
+  const { prompt, imageUrl, model, timeout = 2e5, onLog } = params;
   return runMindstudioCli(
     [
       "analyze-image",
@@ -3094,21 +3193,15 @@ async function analyzeImage(params) {
       "--image-url",
       imageUrl,
       "--vision-model-override",
-      JSON.stringify(VISION_MODEL_OVERRIDE)
+      JSON.stringify({ model })
     ],
     { outputKey: "analysis", timeout, onLog }
   );
 }
-var VISION_MODEL, VISION_MODEL_OVERRIDE;
 var init_analyzeImage = __esm({
   "src/subagents/common/analyzeImage.ts"() {
     "use strict";
     init_runMindstudioCli();
-    VISION_MODEL = "claude-4-6-sonnet";
-    VISION_MODEL_OVERRIDE = {
-      model: VISION_MODEL,
-      config: { thinkingBudget: "off" }
-    };
   }
 });
 
@@ -3130,13 +3223,14 @@ ${TEXT_WRAP_DISCLAIMER}`;
   return p;
 }
 async function streamScreenshotAnalysis(opts) {
-  const { url, prompt, styleMap, onLog } = opts;
+  const { url, prompt, styleMap, onLog, model } = opts;
   onLog?.(JSON.stringify({ url, analysis: null }));
   const analysisPrompt = buildScreenshotAnalysisPrompt({ prompt, styleMap });
   let accumulated = "";
   const analysis = await analyzeImage({
     prompt: analysisPrompt,
     imageUrl: url,
+    model,
     onLog: (chunk) => {
       accumulated += chunk;
       onLog?.(JSON.stringify({ url, analysis: accumulated }));
@@ -3148,12 +3242,14 @@ async function captureAndAnalyzeScreenshot(promptOrOptions) {
   let prompt;
   let existingUrl;
   let onLog;
+  let model;
   let path13;
   if (typeof promptOrOptions === "object" && promptOrOptions !== null) {
     prompt = promptOrOptions.prompt;
     existingUrl = promptOrOptions.imageUrl;
     path13 = promptOrOptions.path;
     onLog = promptOrOptions.onLog;
+    model = promptOrOptions.model;
   } else {
     prompt = promptOrOptions;
   }
@@ -3178,11 +3274,17 @@ async function captureAndAnalyzeScreenshot(promptOrOptions) {
   if (prompt === false) {
     return url;
   }
+  if (!model) {
+    throw new Error(
+      "captureAndAnalyzeScreenshot: `model` is required when analysis is enabled"
+    );
+  }
   return streamScreenshotAnalysis({
     url,
     prompt: prompt || void 0,
     styleMap,
-    onLog
+    onLog,
+    model
   });
 }
 var SCREENSHOT_ANALYSIS_PROMPT, TEXT_WRAP_DISCLAIMER;
@@ -4085,7 +4187,7 @@ var init_browserAutomation = __esm({
     init_browserLock();
     init_screenshot();
     init_runMindstudioCli();
-    init_analyzeImage();
+    init_surfaces();
     init_logger();
     log6 = createLogger("browser-automation");
     browserAutomationTool = {
@@ -4135,7 +4237,12 @@ var init_browserAutomation = __esm({
                 try {
                   return await captureAndAnalyzeScreenshot({
                     path: _input.path,
-                    onLog
+                    onLog,
+                    model: resolveModel(
+                      "imageAnalysis",
+                      context.models,
+                      context.model
+                    )
                   });
                 } catch (err) {
                   return `Error taking screenshot: ${err.message}`;
@@ -4144,7 +4251,7 @@ var init_browserAutomation = __esm({
               return `Error: unknown local tool "${name}"`;
             },
             apiConfig: context.apiConfig,
-            model: context.models?.browserAutomation ?? context.model,
+            model: resolveModel("browserAutomation", context.models, context.model),
             subAgentId: "browserAutomation",
             signal: context.signal,
             parentToolId: context.toolCallId,
@@ -4162,6 +4269,13 @@ var init_browserAutomation = __esm({
                     (s) => s.command === "screenshotViewport" && s.result?.url
                   );
                   if (screenshotSteps.length > 0) {
+                    const visionOverride = {
+                      model: resolveModel(
+                        "imageAnalysis",
+                        context.models,
+                        context.model
+                      )
+                    };
                     const batchInput = screenshotSteps.map((s) => ({
                       stepType: "analyzeImage",
                       step: {
@@ -4169,7 +4283,7 @@ var init_browserAutomation = __esm({
                         prompt: buildScreenshotAnalysisPrompt({
                           styleMap: s.result.styleMap
                         }),
-                        visionModelOverride: VISION_MODEL_OVERRIDE
+                        visionModelOverride: visionOverride
                       }
                     }));
                     const batchResult = await runMindstudioCli(
@@ -4224,6 +4338,7 @@ var init_screenshot2 = __esm({
     init_screenshot();
     init_browserLock();
     init_browserAutomation();
+    init_surfaces();
     screenshotTool = {
       clearable: true,
       definition: {
@@ -4257,7 +4372,8 @@ var init_screenshot2 = __esm({
             return await captureAndAnalyzeScreenshot({
               prompt: input.prompt,
               imageUrl: input.imageUrl,
-              onLog: context?.onLog
+              onLog: context?.onLog,
+              model: resolveModel("imageAnalysis", context?.models, context?.model)
             });
           }
           if (input.instructions && context) {
@@ -4279,7 +4395,8 @@ var init_screenshot2 = __esm({
               url,
               prompt: input.prompt,
               styleMap,
-              onLog: context?.onLog
+              onLog: context?.onLog,
+              model: resolveModel("imageAnalysis", context?.models, context?.model)
             });
           }
           const release = await acquireBrowserLock();
@@ -4287,7 +4404,8 @@ var init_screenshot2 = __esm({
             return await captureAndAnalyzeScreenshot({
               prompt: input.prompt,
               path: input.path,
-              onLog: context?.onLog
+              onLog: context?.onLog,
+              model: resolveModel("imageAnalysis", context?.models, context?.model)
             });
           } finally {
             release();
@@ -4478,7 +4596,7 @@ __export(analyzeDesign_exports, {
   definition: () => definition3,
   execute: () => execute3
 });
-async function execute3(input, onLog) {
+async function execute3(input, onLog, context) {
   const url = input.url;
   const analysisPrompt = input.prompt || DESIGN_REFERENCE_PROMPT;
   const isImageUrl = /\.(png|jpe?g|webp|gif|svg|avif)(\?|$)/i.test(url);
@@ -4511,7 +4629,8 @@ async function execute3(input, onLog) {
   const analysis = await analyzeImage({
     prompt: analysisPrompt,
     imageUrl,
-    onLog
+    onLog,
+    model: resolveModel("imageAnalysis", context?.models, context?.model)
   });
   return JSON.stringify({ url: imageUrl, analysis });
 }
@@ -4521,6 +4640,7 @@ var init_analyzeDesign = __esm({
     "use strict";
     init_runMindstudioCli();
     init_analyzeImage();
+    init_surfaces();
     DESIGN_REFERENCE_PROMPT = `
 You are analyzing a screenshot of a real website or app for a designer's personal technique/inspiration reference notes.
 
@@ -4572,7 +4692,7 @@ __export(analyzeImage_exports, {
   definition: () => definition4,
   execute: () => execute4
 });
-async function execute4(input, onLog) {
+async function execute4(input, onLog, context) {
   const imageUrl = input.imageUrl;
   const prompt = buildScreenshotAnalysisPrompt({
     prompt: input.prompt
@@ -4580,7 +4700,8 @@ async function execute4(input, onLog) {
   const analysis = await analyzeImage({
     prompt,
     imageUrl,
-    onLog
+    onLog,
+    model: resolveModel("imageAnalysis", context?.models, context?.model)
   });
   return JSON.stringify({ url: imageUrl, analysis });
 }
@@ -4590,6 +4711,7 @@ var init_analyzeImage2 = __esm({
     "use strict";
     init_analyzeImage();
     init_screenshot();
+    init_surfaces();
     definition4 = {
       clearable: true,
       name: "analyzeImage",
@@ -4639,7 +4761,8 @@ async function execute5(input, onLog, context) {
         url,
         prompt: input.prompt,
         styleMap,
-        onLog
+        onLog,
+        model: resolveModel("imageAnalysis", context?.models, context?.model)
       });
     } catch (err) {
       return `Error taking interactive screenshot: ${err.message}`;
@@ -4650,7 +4773,8 @@ async function execute5(input, onLog, context) {
     return await captureAndAnalyzeScreenshot({
       prompt: input.prompt,
       path: input.path,
-      onLog
+      onLog,
+      model: resolveModel("imageAnalysis", context?.models, context?.model)
     });
   } catch (err) {
     return `Error taking screenshot: ${err.message}`;
@@ -4665,6 +4789,7 @@ var init_screenshot3 = __esm({
     init_screenshot();
     init_browserLock();
     init_browserAutomation();
+    init_surfaces();
     definition5 = {
       clearable: true,
       name: "screenshot",
@@ -4692,7 +4817,7 @@ var init_screenshot3 = __esm({
 
 // src/subagents/designExpert/tools/images/enhancePrompt.ts
 async function enhanceImagePrompt(params) {
-  const { brief, width, height, transparentBackground, onLog } = params;
+  const { brief, width, height, transparentBackground, onLog, model } = params;
   const contextParts = [
     `Dimensions: ${width}x${height}${width > height ? " (landscape)" : width < height ? " (portrait)" : " (square)"}`
   ];
@@ -4717,23 +4842,18 @@ ${brief}
       "--message",
       message,
       "--model-override",
-      JSON.stringify(MODEL_OVERRIDE)
+      JSON.stringify({ model, config: { reasoning: "false" } })
     ],
     { outputKey: "content", timeout: 6e4, onLog, caller: "designExpert" }
   );
   return enhanced.trim();
 }
-var ENHANCE_MODEL, MODEL_OVERRIDE, SYSTEM_PROMPT;
+var SYSTEM_PROMPT;
 var init_enhancePrompt = __esm({
   "src/subagents/designExpert/tools/images/enhancePrompt.ts"() {
     "use strict";
     init_runMindstudioCli();
     init_assets();
-    ENHANCE_MODEL = "claude-4-6-sonnet";
-    MODEL_OVERRIDE = {
-      model: ENHANCE_MODEL,
-      config: { reasoning: "false" }
-    };
     SYSTEM_PROMPT = readAsset(
       "subagents/designExpert/tools/images/enhance-image-prompt.md"
     );
@@ -4742,7 +4862,15 @@ var init_enhancePrompt = __esm({
 
 // src/subagents/designExpert/tools/images/imageGenerator.ts
 async function generateImageAssets(opts) {
-  const { prompts, sourceImages, transparentBackground, onLog } = opts;
+  const {
+    prompts,
+    sourceImages,
+    transparentBackground,
+    onLog,
+    imageGenerationModel: genModel,
+    imageAnalysisModel,
+    imagePromptEnhancerModel
+  } = opts;
   const width = opts.width || 2048;
   const height = opts.height || 2048;
   const config = { width, height };
@@ -4757,7 +4885,8 @@ async function generateImageAssets(opts) {
         width,
         height,
         transparentBackground,
-        onLog
+        onLog,
+        model: imagePromptEnhancerModel
       })
     )
   );
@@ -4766,7 +4895,7 @@ async function generateImageAssets(opts) {
     const step = JSON.stringify({
       prompt: enhancedPrompts[0],
       imageModelOverride: {
-        model: "seedream-4.5",
+        model: genModel,
         config
       }
     });
@@ -4785,7 +4914,7 @@ async function generateImageAssets(opts) {
       step: {
         prompt,
         imageModelOverride: {
-          model: "seedream-4.5",
+          model: genModel,
           config
         }
       }
@@ -4837,7 +4966,8 @@ async function generateImageAssets(opts) {
       const analysis = await analyzeImage({
         prompt: ANALYZE_PROMPT,
         imageUrl: url,
-        onLog
+        onLog,
+        model: imageAnalysisModel
       });
       return {
         url,
@@ -4868,13 +4998,28 @@ __export(generateImages_exports, {
   definition: () => definition6,
   execute: () => execute6
 });
-async function execute6(input, onLog) {
+async function execute6(input, onLog, context) {
   return generateImageAssets({
     prompts: input.prompts,
     width: input.width,
     height: input.height,
     transparentBackground: input.transparentBackground,
-    onLog
+    onLog,
+    imageGenerationModel: resolveModel(
+      "imageGeneration",
+      context?.models,
+      context?.model
+    ),
+    imageAnalysisModel: resolveModel(
+      "imageAnalysis",
+      context?.models,
+      context?.model
+    ),
+    imagePromptEnhancerModel: resolveModel(
+      "imagePromptEnhancer",
+      context?.models,
+      context?.model
+    )
   });
 }
 var definition6;
@@ -4882,6 +5027,7 @@ var init_generateImages = __esm({
   "src/subagents/designExpert/tools/images/generateImages.ts"() {
     "use strict";
     init_imageGenerator();
+    init_surfaces();
     definition6 = {
       clearable: false,
       name: "generateImages",
@@ -4921,14 +5067,29 @@ __export(editImages_exports, {
   definition: () => definition7,
   execute: () => execute7
 });
-async function execute7(input, onLog) {
+async function execute7(input, onLog, context) {
   return generateImageAssets({
     prompts: input.prompts,
     sourceImages: input.sourceImages,
     width: input.width,
     height: input.height,
     transparentBackground: input.transparentBackground,
-    onLog
+    onLog,
+    imageGenerationModel: resolveModel(
+      "imageGeneration",
+      context?.models,
+      context?.model
+    ),
+    imageAnalysisModel: resolveModel(
+      "imageAnalysis",
+      context?.models,
+      context?.model
+    ),
+    imagePromptEnhancerModel: resolveModel(
+      "imagePromptEnhancer",
+      context?.models,
+      context?.model
+    )
   });
 }
 var definition7;
@@ -4936,6 +5097,7 @@ var init_editImages = __esm({
   "src/subagents/designExpert/tools/images/editImages.ts"() {
     "use strict";
     init_imageGenerator();
+    init_surfaces();
     definition7 = {
       clearable: false,
       name: "editImages",
@@ -5474,6 +5636,7 @@ var init_designExpert = __esm({
     init_tools2();
     init_prompt3();
     init_history();
+    init_surfaces();
     DESCRIPTION = `
 Visual design expert. Describe the situation and what you need \u2014 the agent decides what to deliver. It reads the spec files automatically. Include relevant user requirements and context it can't get from the spec, but do not list specific deliverables or tell it how to do its job. Do not suggest implementation details or ideas - only relay what is needed.
 `.trim();
@@ -5522,7 +5685,7 @@ Visual design expert. Describe the situation and what you need \u2014 the agent
             );
           },
           apiConfig: context.apiConfig,
-          model: context.models?.visualDesignExpert ?? context.model,
+          model: resolveModel("visualDesignExpert", context.models, context.model),
           subAgentId: "visualDesignExpert",
           signal: context.signal,
           parentToolId: context.toolCallId,
@@ -5740,6 +5903,7 @@ var init_productVision = __esm({
     init_executor();
     init_prompt4();
     init_history();
+    init_surfaces();
     productVisionTool = {
       clearable: false,
       definition: {
@@ -5779,7 +5943,7 @@ var init_productVision = __esm({
             return executeVisionTool(name, input2, childCtx);
           },
           apiConfig: context.apiConfig,
-          model: context.models?.productVision ?? context.model,
+          model: resolveModel("productVision", context.models, context.model),
           subAgentId: "productVision",
           signal: context.signal,
           parentToolId: context.toolCallId,
@@ -5870,6 +6034,7 @@ var init_codeSanityCheck = __esm({
     init_context();
     init_tools6();
     init_tools5();
+    init_surfaces();
     BASE_PROMPT3 = readAsset("subagents/codeSanityCheck", "prompt.md");
     codeSanityCheckTool = {
       clearable: false,
@@ -5905,7 +6070,7 @@ var init_codeSanityCheck = __esm({
           externalTools: /* @__PURE__ */ new Set(),
           executeTool: (name, toolInput) => executeTool(name, toolInput, context),
           apiConfig: context.apiConfig,
-          model: context.models?.codeSanityCheck ?? context.model,
+          model: resolveModel("codeSanityCheck", context.models, context.model),
           subAgentId: "codeSanityCheck",
           signal: context.signal,
           parentToolId: context.toolCallId,
@@ -6869,7 +7034,7 @@ async function runTurn(params) {
         onEvent({ type: "tool_input_delta", id, name, result: content });
       }
     }
-    const parentModel = state.models?.parent ?? model;
+    const parentModel = resolveModel("parent", state.models, model);
     try {
       for await (const event of streamChatWithRetry(
         {
@@ -7213,7 +7378,10 @@ async function runTurn(params) {
           isError: r.isError
         });
         if (!r.isError && BRAND_TRIGGERING_TOOLS.has(tc.name)) {
-          triggerBrandExtraction(apiConfig);
+          triggerBrandExtraction(
+            apiConfig,
+            resolveModel("brandExtractor", state.models, model)
+          );
         }
         return r;
       })
@@ -7270,6 +7438,7 @@ var init_agent = __esm({
     init_tools6();
     init_sentinel();
     init_trigger2();
+    init_surfaces();
     log10 = createLogger("agent");
     BRAND_TRIGGERING_TOOLS = /* @__PURE__ */ new Set(["writeSpec", "editSpec"]);
     EXTERNAL_TOOLS = /* @__PURE__ */ new Set([
@@ -7731,6 +7900,7 @@ var init_headless = __esm({
     init_lsp();
     init_agent();
     init_session();
+    init_surfaces();
     init_toolRegistry();
     init_attachments();
     init_planFile();
@@ -7811,12 +7981,14 @@ var init_headless = __esm({
           this.emit("session_restored", {
             messageCount: this.state.messages.length,
             ...this.state.models && { models: this.state.models },
+            modelSurfaces: MODEL_SURFACES,
+            allowedModelsByType: ALLOWED_MODELS_BY_TYPE,
             ...this.queueFields()
           });
         }
         triggerBrandExtraction(
           this.config,
-          this.state.models?.brandExtractor ?? this.opts.model
+          resolveModel("brandExtractor", this.state.models, this.opts.model)
         );
         this.toolRegistry.onEvent = this.onEvent;
         setCompactionListener((event) => {
@@ -8494,6 +8666,8 @@ var init_headless = __esm({
             running: this.running,
             ...this.running && this.currentRequestId ? { currentRequestId: this.currentRequestId } : {},
             ...this.state.models && { models: this.state.models },
+            modelSurfaces: MODEL_SURFACES,
+            allowedModelsByType: ALLOWED_MODELS_BY_TYPE,
             ...this.queueFields()
           }));
           return;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mindstudio-ai/remy",
-  "version": "0.1.178",
+  "version": "0.1.179",
   "description": "MindStudio coding agent",
   "repository": {
     "type": "git",