feat: Add reasoning and capability controls for OpenAI Compatible models

.changeset/disable-questions-yolo-mode.md

@@ -3,6 +3,7 @@
 ---
 
 Disable ask_followup_question tool when yolo mode is enabled to prevent the agent from asking itself questions and auto-answering them. Applied to:
+
 - XML tool descriptions (system prompt)
 - Native tool filtering
 - Tool execution (returns error message if model still tries to use the tool from conversation history)

.changeset/openai-compatible-settings-improvements.md

@@ -0,0 +1,10 @@
+---
+"kilo-code": minor
+---
+
+Add reasoning and capability controls for OpenAI Compatible models
+
+- Added checkboxes for 'Supports Reasoning', 'Supports Function Calling', and 'Supports Computer Use' to the OpenAI Compatible settings UI.
+- Compacted the capability checkboxes into a 2-column grid layout with tooltip-only descriptions.
+- Updated OpenAiHandler to inject the 'thinking' parameter when reasoning is enabled and the model supports it.
+- Gated tool inclusion based on the 'supportsNativeTools' flag.

.github/workflows/code-qa.yml

@@ -186,6 +186,7 @@ jobs:
                     libxkbfile-dev \
                     pkg-config \
                     build-essential \
+                    libkrb5-dev \
                     python3
             - name: Turbo cache setup
               uses: actions/cache@v4

.github/workflows/marketplace-publish.yml

@@ -201,6 +201,7 @@ jobs:
                     libxkbfile-dev \
                     pkg-config \
                     build-essential \
+                    libkrb5-dev \
                     python3
             - name: Turbo cache setup
               uses: actions/cache@v4

AGENTS.md

@@ -14,6 +14,7 @@ This is a pnpm monorepo using Turbo for task orchestration:
 - **`apps/`** - E2E tests, Storybook, docs
 
 Key source directories:
+
 - `src/api/providers/` - AI provider implementations (50+ providers)
 - `src/core/tools/` - Tool implementations (ReadFile, ApplyDiff, ExecuteCommand, etc.)
 - `src/services/` - Services (MCP, browser, checkpoints, code-index)
@@ -67,11 +68,13 @@ Kilo Code is a fork of [Roo Code](https://github.com/RooVetGit/Roo-Code). We per
 To minimize merge conflicts when syncing with upstream, mark Kilo Code-specific changes in shared code with `kilocode_change` comments.
 
 **Single line:**
+
 ```typescript
 const value = 42 // kilocode_change
 ```
 
 **Multi-line:**
+
 ```typescript
 // kilocode_change start
 const foo = 1
@@ -80,6 +83,7 @@ const bar = 2
 ```
 
 **New files:**
+
 ```typescript
 // kilocode_change - new file
 ```

CHANGELOG.md

@@ -1,5 +1,9 @@
 # kilo-code
 
+## 4.143.21
+
+- Fix JetBrains plugin build by handling missing gradle.properties in CI check script. Thanks @janpaul123, @kevinvandijk, @chrarnoldus, @hassoncs, @markijbema, @catrielmuller! - Fixes #3271
+
 ## 4.143.2
 
 ### Patch Changes

apps/kilocode-docs/docs/advanced-usage/agent-manager.md

@@ -90,20 +90,22 @@ The Agent Manager requires proper authentication for full functionality, includi
 ### Supported Authentication Methods
 
 1. **Kilo Code Extension (Recommended)**
-   - Sign in through the extension settings
-   - Provides seamless authentication for the Agent Manager
-   - Enables session syncing and cloud features
+
+    - Sign in through the extension settings
+    - Provides seamless authentication for the Agent Manager
+    - Enables session syncing and cloud features
 
 2. **CLI with Kilo Code Provider**
-   - Use the CLI configured with `kilocode` as the provider
-   - Run `kilocode config` to set up authentication
-   - See [CLI setup](/cli) for details
+    - Use the CLI configured with `kilocode` as the provider
+    - Run `kilocode config` to set up authentication
+    - See [CLI setup](/cli) for details
 
 ### BYOK Limitations
 
 **Important:** Bring Your Own Key (BYOK) is not yet supported with the Agent Manager.
 
 If you're using BYOK with providers like Anthropic, OpenAI, or OpenRouter:
+
 - The Agent Manager will not have access to cloud-synced sessions
 - Session syncing features will be unavailable
 - You must use one of the supported authentication methods above for full functionality

apps/kilocode-docs/docs/providers/bedrock.md

@@ -4,7 +4,7 @@ sidebar_label: AWS Bedrock
 
 # Using AWS Bedrock With Kilo Code
 
-Kilo Code supports accessing models through Amazon Bedrock, a fully managed service that makes a selection of high-performing foundation models (FMs) from leading AI companies available via a single API. This provider connects directly to AWS Bedrock and authenticates with the provided credentials. 
+Kilo Code supports accessing models through Amazon Bedrock, a fully managed service that makes a selection of high-performing foundation models (FMs) from leading AI companies available via a single API. This provider connects directly to AWS Bedrock and authenticates with the provided credentials.
 
 **Website:** [https://aws.amazon.com/bedrock/](https://aws.amazon.com/bedrock/)
 

cli/docs/DEVELOPMENT.md

@@ -7,25 +7,29 @@ We use `pnpm` for package management. Please make sure `pnpm` is installed.
 The CLI is currently built by bundling the extension core and replacing the vscode rendering parts with a cli rendering engine. To _develop_ on the CLI you need to follow a few steps:
 
 1. Install dependencies from the root workspace folder:
-   ```bash
-   pnpm install
-   ```
+
+    ```bash
+    pnpm install
+    ```
 
 2. Set up your environment file. Copy the sample and configure your API keys:
-   ```bash
-   cp .env.sample cli/dist/.env
-   # Edit cli/dist/.env with your API keys
-   ```
+
+    ```bash
+    cp .env.sample cli/dist/.env
+    # Edit cli/dist/.env with your API keys
+    ```
 
 3. Build the extension core from the root workspace folder:
-   ```bash
-   pnpm cli:bundle
-   ```
+
+    ```bash
+    pnpm cli:bundle
+    ```
 
 4. Change into the cli folder:
-   ```bash
-   cd ./cli
-   ```
+
+    ```bash
+    cd ./cli
+    ```
 
 5. Build & run the extension by running `pnpm start:dev`. If you want to use the CLI to work on its own code, you can run `pnpm start:dev -w ../` which will start it within the root workspace folder.
 

docs/context-window-autofill.md

@@ -0,0 +1,45 @@
+# Context Window Auto-fill Feature
+
+## Objective
+
+Implement an auto-fill feature for the context window and other model capabilities in the OpenAI Compatible settings.
+
+## Changes
+
+### Backend
+
+1.  **`src/shared/WebviewMessage.ts`**:
+
+    - Added `requestOpenAiModelInfo` to `WebviewMessage` type.
+    - This message allows the frontend to request model information based on the selected model ID.
+
+2.  **`src/shared/ExtensionMessage.ts`**:
+
+    - Added `openAiModelInfo` to `ExtensionMessage` type.
+    - This property carriers the `ModelInfo` payload back to the frontend.
+
+3.  **`src/api/providers/openai.ts`**:
+
+    - Imported known model maps (`openAiNativeModels`, `anthropicModels`, etc.) from `@roo-code/types`.
+    - Added `getOpenAiModelInfo(modelId: string)` helper function.
+    - This function iterates through known model maps to find and return the `ModelInfo` for a given model ID.
+
+4.  **`src/core/webview/webviewMessageHandler.ts`**:
+    - Added a handler for `requestOpenAiModelInfo`.
+    - It calls `getOpenAiModelInfo` and sends back an `openAiModelInfo` message with the result.
+
+### Frontend
+
+1.  **`webview-ui/src/i18n/locales/en/settings.json`**:
+
+    - Added `"autoFill": "Auto-fill"` translation key.
+
+2.  **`webview-ui/src/components/settings/providers/OpenAICompatible.tsx`**:
+    - Imported `vscode` utility for message passing.
+    - Implemented `handleAutoFill` function that sends `requestOpenAiModelInfo`.
+    - Added a listener in `onMessage` to handle `openAiModelInfo` response and update `openAiCustomModelInfo` state.
+    - Added an "Auto-fill" button in the "Model Capabilities" section header.
+
+## Verification
+
+- Ran `pnpm check-types` successfully, confirming type safety across the monorepo.

jetbrains/scripts/check-dependencies.js

@@ -309,6 +309,21 @@ function checkBuildSystem() {
 	const gradlew = path.join(pluginDir, process.platform === "win32" ? "gradlew.bat" : "gradlew")
 	const buildGradle = path.join(pluginDir, "build.gradle.kts")
 	const gradleProps = path.join(pluginDir, "gradle.properties")
+	const gradlePropsTemplate = path.join(pluginDir, "gradle.properties.template")
+
+	// Auto-generate gradle.properties from template if missing
+	if (!fs.existsSync(gradleProps) && fs.existsSync(gradlePropsTemplate)) {
+		try {
+			printWarning("gradle.properties is missing, generating from template...")
+			let content = fs.readFileSync(gradlePropsTemplate, "utf8")
+			// Use a default version for CI check - strict version sync happens later via sync:version
+			content = content.replace("{{VERSION}}", "0.0.0-dev")
+			fs.writeFileSync(gradleProps, content)
+			printFix("Generated gradle.properties from template")
+		} catch (error) {
+			printError(`Failed to generate gradle.properties: ${error.message}`)
+		}
+	}
 
 	if (fs.existsSync(gradlew) && fs.existsSync(buildGradle) && fs.existsSync(gradleProps)) {
 		printSuccess("Gradle build system is configured")

src/api/providers/openai.ts

@@ -8,6 +8,14 @@ import {
 	openAiModelInfoSaneDefaults,
 	DEEP_SEEK_DEFAULT_TEMPERATURE,
 	OPENAI_AZURE_AI_INFERENCE_PATH,
+	openAiNativeModels,
+	anthropicModels,
+	geminiModels,
+	mistralModels,
+	deepSeekModels,
+	qwenCodeModels,
+	vertexModels,
+	bedrockModels,
 } from "@roo-code/types"
 
 import type { ApiHandlerOptions } from "../../shared/api"
@@ -164,11 +172,17 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				stream: true as const,
 				...(isGrokXAI ? {} : { stream_options: { include_usage: true } }),
 				...(reasoning && reasoning),
-				...(metadata?.tools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
-				...(metadata?.tool_choice && { tool_choice: metadata.tool_choice }),
-				...(metadata?.toolProtocol === "native" && {
-					parallel_tool_calls: metadata.parallelToolCalls ?? false,
-				}),
+				...((this.options.enableReasoningEffort && modelInfo.supportsReasoningBinary
+					? { thinking: { type: "enabled" } }
+					: {}) as any),
+				...(metadata?.tools &&
+					modelInfo.supportsNativeTools !== false && { tools: this.convertToolsForOpenAI(metadata.tools) }),
+				...(metadata?.tool_choice &&
+					modelInfo.supportsNativeTools !== false && { tool_choice: metadata.tool_choice }),
+				...(metadata?.toolProtocol === "native" &&
+					modelInfo.supportsNativeTools !== false && {
+						parallel_tool_calls: metadata.parallelToolCalls ?? false,
+					}),
 			}
 
 			// Add max_tokens if needed
@@ -251,11 +265,17 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 					: enabledLegacyFormat
 						? [systemMessage, ...convertToSimpleMessages(messages)]
 						: [systemMessage, ...convertToOpenAiMessages(messages)],
-				...(metadata?.tools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
-				...(metadata?.tool_choice && { tool_choice: metadata.tool_choice }),
-				...(metadata?.toolProtocol === "native" && {
-					parallel_tool_calls: metadata.parallelToolCalls ?? false,
-				}),
+				...(metadata?.tools &&
+					modelInfo.supportsNativeTools !== false && { tools: this.convertToolsForOpenAI(metadata.tools) }),
+				...(metadata?.tool_choice &&
+					modelInfo.supportsNativeTools !== false && { tool_choice: metadata.tool_choice }),
+				...(metadata?.toolProtocol === "native" &&
+					modelInfo.supportsNativeTools !== false && {
+						parallel_tool_calls: metadata.parallelToolCalls ?? false,
+					}),
+				...((this.options.enableReasoningEffort && modelInfo.supportsReasoningBinary
+					? { thinking: { type: "enabled" } }
+					: {}) as any),
 			}
 
 			// Add max_tokens if needed
@@ -387,11 +407,17 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				...(isGrokXAI ? {} : { stream_options: { include_usage: true } }),
 				reasoning_effort: modelInfo.reasoningEffort as "low" | "medium" | "high" | undefined,
 				temperature: undefined,
-				...(metadata?.tools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
-				...(metadata?.tool_choice && { tool_choice: metadata.tool_choice }),
-				...(metadata?.toolProtocol === "native" && {
-					parallel_tool_calls: metadata.parallelToolCalls ?? false,
-				}),
+				...(metadata?.tools &&
+					modelInfo.supportsNativeTools !== false && { tools: this.convertToolsForOpenAI(metadata.tools) }),
+				...(metadata?.tool_choice &&
+					modelInfo.supportsNativeTools !== false && { tool_choice: metadata.tool_choice }),
+				...(metadata?.toolProtocol === "native" &&
+					modelInfo.supportsNativeTools !== false && {
+						parallel_tool_calls: metadata.parallelToolCalls ?? false,
+					}),
+				...((this.options.enableReasoningEffort && modelInfo.supportsReasoningBinary
+					? { thinking: { type: "enabled" } }
+					: {}) as any),
 			}
 
 			// O3 family models do not support the deprecated max_tokens parameter
@@ -422,11 +448,17 @@ export class OpenAiHandler extends BaseProvider implements SingleCompletionHandl
 				],
 				reasoning_effort: modelInfo.reasoningEffort as "low" | "medium" | "high" | undefined,
 				temperature: undefined,
-				...(metadata?.tools && { tools: this.convertToolsForOpenAI(metadata.tools) }),
-				...(metadata?.tool_choice && { tool_choice: metadata.tool_choice }),
-				...(metadata?.toolProtocol === "native" && {
-					parallel_tool_calls: metadata.parallelToolCalls ?? false,
-				}),
+				...(metadata?.tools &&
+					modelInfo.supportsNativeTools !== false && { tools: this.convertToolsForOpenAI(metadata.tools) }),
+				...(metadata?.tool_choice &&
+					modelInfo.supportsNativeTools !== false && { tool_choice: metadata.tool_choice }),
+				...(metadata?.toolProtocol === "native" &&
+					modelInfo.supportsNativeTools !== false && {
+						parallel_tool_calls: metadata.parallelToolCalls ?? false,
+					}),
+				...((this.options.enableReasoningEffort && modelInfo.supportsReasoningBinary
+					? { thinking: { type: "enabled" } }
+					: {}) as any),
 			}
 
 			// O3 family models do not support the deprecated max_tokens parameter
@@ -574,3 +606,60 @@ export async function getOpenAiModels(baseUrl?: string, apiKey?: string, openAiH
 		return []
 	}
 }
+
+export function getOpenAiModelInfo(modelId: string): ModelInfo | undefined {
+	const models: Record<string, ModelInfo>[] = [
+		openAiNativeModels,
+		anthropicModels,
+		geminiModels,
+		mistralModels,
+		deepSeekModels,
+		qwenCodeModels,
+		vertexModels,
+		bedrockModels,
+	]
+
+	// Helper function to sanitize and return model info
+	const sanitizeModelInfo = (info: ModelInfo): ModelInfo => {
+		if (info.tiers) {
+			return {
+				...info,
+				tiers: info.tiers.map((tier) => ({
+					...tier,
+					// Replace Infinity/null with Number.MAX_SAFE_INTEGER (essentially unlimited)
+					contextWindow:
+						tier.contextWindow === Infinity || tier.contextWindow === null
+							? Number.MAX_SAFE_INTEGER
+							: tier.contextWindow,
+				})),
+			}
+		}
+		return info
+	}
+
+	// Try exact match first
+	for (const modelMap of models) {
+		if (modelId in modelMap) {
+			return sanitizeModelInfo(modelMap[modelId])
+		}
+	}
+
+	// Normalize search ID: remove provider prefix (e.g., "google/", "anthropic/") and convert to lowercase
+	const normalizedSearchId = modelId.replace(/^[a-z-]+\//i, "").toLowerCase()
+
+	// Try fuzzy matching: find models where the key contains the normalized search ID
+	// or where the normalized search ID contains the model key's base name
+	for (const modelMap of models) {
+		const keys = Object.keys(modelMap)
+		for (const key of keys) {
+			const normalizedKey = key.toLowerCase()
+			// Check if key contains search term or search term contains key's base (without date suffix)
+			const keyBase = normalizedKey.replace(/-\d{8}$/, "") // Remove date suffixes like -20241022
+			if (normalizedKey.includes(normalizedSearchId) || normalizedSearchId.includes(keyBase)) {
+				return sanitizeModelInfo(modelMap[key])
+			}
+		}
+	}
+
+	return undefined
+}