Merge branch 'main' into static-context-ide-readfile

Author: sestinj

core/config/markdown/loadMarkdownRules.ts

@@ -79,7 +79,13 @@ export async function loadMarkdownRules(ide: IDE): Promise<{
             uriType: "file",
             fileUri: file.path,
           });
-          rules.push({ ...rule, source: "rules-block", sourceFile: file.path });
+          if (!rule.invokable) {
+            rules.push({
+              ...rule,
+              source: "rules-block",
+              sourceFile: file.path,
+            });
+          }
         } catch (e) {
           errors.push({
             fatal: false,

core/index.d.ts

@@ -359,11 +359,15 @@ export interface ToolResultChatMessage {
   role: "tool";
   content: string;
   toolCallId: string;
+  /** Arbitrary per-message metadata (IDs, provider-specific info, etc.) */
+  metadata?: Record<string, unknown>;
 }
 
 export interface UserChatMessage {
   role: "user";
   content: MessageContent;
+  /** Arbitrary per-message metadata (IDs, provider-specific info, etc.) */
+  metadata?: Record<string, unknown>;
 }
 
 export interface ThinkingChatMessage {
@@ -372,6 +376,12 @@ export interface ThinkingChatMessage {
   signature?: string;
   redactedThinking?: string;
   toolCalls?: ToolCallDelta[];
+  reasoning_details?: {
+    signature?: string;
+    [key: string]: any;
+  }[];
+  /** Arbitrary per-message metadata (IDs, provider-specific info, etc.) */
+  metadata?: Record<string, unknown>;
 }
 
 /**
@@ -400,11 +410,15 @@ export interface AssistantChatMessage {
   content: MessageContent;
   toolCalls?: ToolCallDelta[];
   usage?: Usage;
+  /** Arbitrary per-message metadata (IDs, provider-specific info, etc.) */
+  metadata?: Record<string, unknown>;
 }
 
 export interface SystemChatMessage {
   role: "system";
   content: string;
+  /** Arbitrary per-message metadata (IDs, provider-specific info, etc.) */
+  metadata?: Record<string, unknown>;
 }
 
 export type ChatMessage =

core/llm/autodetect.ts

@@ -63,6 +63,47 @@ const PROVIDER_HANDLES_TEMPLATING: string[] = [
   "watsonx",
   "nebius",
   "relace",
+  "openrouter",
+  "deepseek",
+  "xAI",
+  "groq",
+  "gemini",
+  "docker",
+  // TODO add these, change to inverted logic so only the ones that need templating are hardcoded
+  // Asksage.ts
+  // Azure.ts
+  // BedrockImport.ts
+  // Cerebras.ts
+  // Cloudflare.ts
+  // CometAPI.ts
+  // CustomLLM.ts
+  // DeepInfra.ts
+  // Fireworks.ts
+  // Flowise.ts
+  // FunctionNetwork.ts
+  // HuggingFaceInferenceAPI.ts
+  // HuggingFaceTEI.ts
+  // HuggingFaceTGI.ts
+  // Inception.ts
+  // Kindo.ts
+  // LlamaCpp.ts
+  // LlamaStack.ts
+  // Llamafile.ts
+  // Mock.ts
+  // Moonshot.ts
+  // NCompass.ts
+  // OVHcloud.ts
+  // Replicate.ts
+  // Scaleway.ts
+  // SiliconFlow.ts
+  // TARS.ts
+  // Test.ts
+  // TextGenWebUI.ts
+  // TransformersJsEmbeddingsProvider.ts
+  // Venice.ts
+  // Vllm.ts
+  // Voyage.ts
+  // etc
 ];
 
 const PROVIDER_SUPPORTS_IMAGES: string[] = [
@@ -105,6 +146,8 @@ const MODEL_SUPPORTS_IMAGES: RegExp[] = [
   /\bgemma-?3(?!n)/, // gemma3 supports vision, but gemma3n doesn't!
   /\b(pali|med)gemma/,
   /qwen(.*)vl/,
+  /mistral-small/,
+  /mistral-medium/,
 ];
 
 function modelSupportsImages(
@@ -142,7 +185,12 @@ function modelSupportsReasoning(
   if (!model) {
     return false;
   }
-  if ("anthropic" === model.underlyingProviderName) {
+  // do not turn reasoning on by default for claude 3 models
+  if (
+    model.model.includes("claude") &&
+    !model.model.includes("-3-") &&
+    !model.model.includes("-3.5-")
+  ) {
     return true;
   }
   if (model.model.includes("deepseek-r")) {