Fix: always send stream flag explicitly in LLM request body

Fluid.xcodeproj/project.pbxproj

@@ -14,6 +14,7 @@
 		7C91B0012F42AA0100C0DEF0 /* HotkeyShortcutTests.swift in Sources */ = {isa = PBXBuildFile; fileRef = 7C91B0022F42AA0100C0DEF0 /* HotkeyShortcutTests.swift */; };
 		7CDB0A2D2F3C4D5600FB7CAD /* DictationE2ETests.swift in Sources */ = {isa = PBXBuildFile; fileRef = 7CDB0A292F3C4D5600FB7CAD /* DictationE2ETests.swift */; };
 		7CDB0A2E2F3C4D5600FB7CAD /* AudioFixtureLoader.swift in Sources */ = {isa = PBXBuildFile; fileRef = 7CDB0A2A2F3C4D5600FB7CAD /* AudioFixtureLoader.swift */; };
+		86CAA2D4EF18433096185602 /* LLMClientRequestBodyTests.swift in Sources */ = {isa = PBXBuildFile; fileRef = 343B29013F4441D6A797D12D /* LLMClientRequestBodyTests.swift */; };
 		7CDB0A2F2F3C4D5600FB7CAD /* dictation_fixture.wav in Resources */ = {isa = PBXBuildFile; fileRef = 7CDB0A2B2F3C4D5600FB7CAD /* dictation_fixture.wav */; };
 		7CDB0A302F3C4D5600FB7CAD /* XCTest.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 7CDB0A2C2F3C4D5600FB7CAD /* XCTest.framework */; };
 		7CE006BD2E80EBE600DDCCD6 /* AppUpdater in Frameworks */ = {isa = PBXBuildFile; productRef = 7CE006BC2E80EBE600DDCCD6 /* AppUpdater */; };
@@ -34,6 +35,7 @@
 		7CDB0A202F3C4D5600FB7CAD /* FluidDictationIntegrationTests.xctest */ = {isa = PBXFileReference; explicitFileType = wrapper.cfbundle; includeInIndex = 0; path = FluidDictationIntegrationTests.xctest; sourceTree = BUILT_PRODUCTS_DIR; };
 		7C91B0022F42AA0100C0DEF0 /* HotkeyShortcutTests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = HotkeyShortcutTests.swift; sourceTree = "<group>"; };
 		7CDB0A292F3C4D5600FB7CAD /* DictationE2ETests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = DictationE2ETests.swift; sourceTree = "<group>"; };
+		343B29013F4441D6A797D12D /* LLMClientRequestBodyTests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = LLMClientRequestBodyTests.swift; sourceTree = "<group>"; };
 		7CDB0A2A2F3C4D5600FB7CAD /* AudioFixtureLoader.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = AudioFixtureLoader.swift; sourceTree = "<group>"; };
 		7CDB0A2B2F3C4D5600FB7CAD /* dictation_fixture.wav */ = {isa = PBXFileReference; lastKnownFileType = audio.wav; path = dictation_fixture.wav; sourceTree = "<group>"; };
 		7CDB0A2C2F3C4D5600FB7CAD /* XCTest.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = XCTest.framework; path = Platforms/MacOSX.platform/Developer/Library/Frameworks/XCTest.framework; sourceTree = DEVELOPER_DIR; };
@@ -104,6 +106,7 @@
 				7CDB0A272F3C4D5600FB7CAD /* Resources */,
 				7CDB0A292F3C4D5600FB7CAD /* DictationE2ETests.swift */,
 				7C91B0022F42AA0100C0DEF0 /* HotkeyShortcutTests.swift */,
+				343B29013F4441D6A797D12D /* LLMClientRequestBodyTests.swift */,
 			);
 			path = FluidDictationIntegrationTests;
 			sourceTree = "<group>";
@@ -258,6 +261,7 @@
 				7CDB0A2E2F3C4D5600FB7CAD /* AudioFixtureLoader.swift in Sources */,
 				7CDB0A2D2F3C4D5600FB7CAD /* DictationE2ETests.swift in Sources */,
 				7C91B0012F42AA0100C0DEF0 /* HotkeyShortcutTests.swift in Sources */,
+				86CAA2D4EF18433096185602 /* LLMClientRequestBodyTests.swift in Sources */,
 			);
 			runOnlyForDeploymentPostprocessing = 0;
 		};

Sources/Fluid/Services/LLMClient.swift

@@ -132,7 +132,7 @@ final class LLMClient {
         var maxRetries: Int = 3
         var retryDelayMs: Int = 200
 
-        // Timeout configuration (nil = use default)
+        /// Timeout configuration (nil = use default)
         var timeoutSeconds: TimeInterval?
 
         // Optional real-time callbacks (for streaming UI updates)
@@ -310,7 +310,7 @@ final class LLMClient {
         request.url?.path.contains("/responses") == true
     }
 
-    private func buildChatCompletionsBody(_ config: Config) -> [String: Any] {
+    func buildChatCompletionsBody(_ config: Config) -> [String: Any] {
         var body: [String: Any] = [
             "model": config.model,
             "messages": config.messages,
@@ -327,10 +327,8 @@ final class LLMClient {
             body["tool_choice"] = "auto"
         }
 
-        // Add streaming flag
-        if config.streaming {
-            body["stream"] = true
-        }
+        // Always send stream explicitly — providers like Ollama treat an absent key as true
+        body["stream"] = config.streaming
 
         // Layer 1: Model-specific parameters (e.g., enable_thinking for Nemotron)
         let modelExtras = ThinkingParserFactory.getExtraParameters(for: config.model)
@@ -355,16 +353,15 @@ final class LLMClient {
         return body
     }
 
-    private func buildResponsesBody(_ config: Config) -> [String: Any] {
+    func buildResponsesBody(_ config: Config) -> [String: Any] {
         var body: [String: Any] = [
             "model": config.model,
             "input": self.responsesInput(from: config.messages),
             "store": false,
         ]
 
-        if config.streaming {
-            body["stream"] = true
-        }
+        // Always send stream explicitly — providers like Ollama treat an absent key as true
+        body["stream"] = config.streaming
 
         if !config.tools.isEmpty {
             body["tools"] = self.responsesTools(from: config.tools)

Tests/FluidDictationIntegrationTests/LLMClientRequestBodyTests.swift

@@ -0,0 +1,46 @@
+@testable import FluidVoice_Debug
+import XCTest
+
+// Regression tests for https://github.com/altic-dev/FluidVoice/issues/295
+// Ollama and compatible OpenAI-format providers treat an absent `stream` key as true.
+// The fix is to always send the key explicitly, whether streaming or not.
+
+@MainActor
+final class LLMClientRequestBodyTests: XCTestCase {
+
+    private func config(streaming: Bool) -> LLMClient.Config {
+        LLMClient.Config(
+            messages: [["role": "user", "content": "hello"]],
+            model: "llama3",
+            baseURL: "http://localhost:11434/v1",
+            apiKey: "",
+            streaming: streaming
+        )
+    }
+
+    // MARK: - Chat Completions endpoint
+
+    func testChatCompletionsBody_streamFalse_keyIsPresentAndFalse() {
+        let body = LLMClient.shared.buildChatCompletionsBody(config(streaming: false))
+        XCTAssertNotNil(body["stream"], "stream key must be present when streaming=false — absent key breaks Ollama-compatible providers")
+        XCTAssertEqual(body["stream"] as? Bool, false)
+    }
+
+    func testChatCompletionsBody_streamTrue_keyIsPresentAndTrue() {
+        let body = LLMClient.shared.buildChatCompletionsBody(config(streaming: true))
+        XCTAssertEqual(body["stream"] as? Bool, true)
+    }
+
+    // MARK: - Responses endpoint
+
+    func testResponsesBody_streamFalse_keyIsPresentAndFalse() {
+        let body = LLMClient.shared.buildResponsesBody(config(streaming: false))
+        XCTAssertNotNil(body["stream"], "stream key must be present when streaming=false")
+        XCTAssertEqual(body["stream"] as? Bool, false)
+    }
+
+    func testResponsesBody_streamTrue_keyIsPresentAndTrue() {
+        let body = LLMClient.shared.buildResponsesBody(config(streaming: true))
+        XCTAssertEqual(body["stream"] as? Bool, true)
+    }
+}