Add OpenRouter dedicated images API support

Fetches /api/v1/images/models in parallel with /models and merges results into the model picker. Image-only models (e.g. Sourceful, Seedream, Flux via this endpoint) were previously invisible since they don't appear in the standard /models endpoint. Models from the images API get usesImagesAPI=true and route through a new generateImageAPIResponse() path in ChatViewModel that POSTs to /api/v1/images with {model, prompt} instead of the chat completions endpoint. The response's b64_json data is decoded and displayed via the existing GeneratedImagesView. Cost is taken directly from the usage.cost field in the images API response (USD per image) via a new rawCostUSD field on ChatResponse.Usage, bypassing the token-based calculateCost() path used for chat models. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-06-29 14:34:56 +02:00
parent 40c5f25517
commit f2949cea3b
5 changed files with 243 additions and 23 deletions
@@ -47,6 +47,7 @@ struct ModelInfo: Identifiable, Codable, Hashable {
        let online: Bool           // Web search
        var imageGeneration: Bool = false  // Image output
        var thinking: Bool = false         // Reasoning/thinking tokens
+        var usesImagesAPI: Bool = false    // OpenRouter dedicated /images endpoint
    }

    struct Architecture: Codable, Hashable {
@@ -132,15 +132,19 @@ struct ChatResponse: Codable {
        let totalTokens: Int
        let cacheCreationInputTokens: Int?
        let cacheReadInputTokens: Int?
+        /// Direct USD cost returned by the images API (bypasses token-based calculation).
+        let rawCostUSD: Double?

-        init(promptTokens: Int, completionTokens: Int, totalTokens: Int, cacheCreationInputTokens: Int? = nil, cacheReadInputTokens: Int? = nil) {
+        init(promptTokens: Int, completionTokens: Int, totalTokens: Int, cacheCreationInputTokens: Int? = nil, cacheReadInputTokens: Int? = nil, rawCostUSD: Double? = nil) {
            self.promptTokens = promptTokens
            self.completionTokens = completionTokens
            self.totalTokens = totalTokens
            self.cacheCreationInputTokens = cacheCreationInputTokens
            self.cacheReadInputTokens = cacheReadInputTokens
+            self.rawCostUSD = rawCostUSD
        }

+        // rawCostUSD is set programmatically, never decoded from API responses
        enum CodingKeys: String, CodingKey {
            case promptTokens = "prompt_tokens"
            case completionTokens = "completion_tokens"
@@ -148,6 +152,16 @@ struct ChatResponse: Codable {
            case cacheCreationInputTokens = "cache_creation_input_tokens"
            case cacheReadInputTokens = "cache_read_input_tokens"
        }
+
+        init(from decoder: Decoder) throws {
+            let c = try decoder.container(keyedBy: CodingKeys.self)
+            promptTokens = try c.decode(Int.self, forKey: .promptTokens)
+            completionTokens = try c.decode(Int.self, forKey: .completionTokens)
+            totalTokens = try c.decode(Int.self, forKey: .totalTokens)
+            cacheCreationInputTokens = try c.decodeIfPresent(Int.self, forKey: .cacheCreationInputTokens)
+            cacheReadInputTokens = try c.decodeIfPresent(Int.self, forKey: .cacheReadInputTokens)
+            rawCostUSD = nil
+        }
    }

    // Custom Codable since ToolCallInfo/generatedImages are not from API directly
@@ -420,6 +420,67 @@ struct ToolResultMessage: Encodable {
    }
 }

+// MARK: - Images API — Model Discovery
+
+struct OpenRouterImageModelsResponse: Codable {
+    let data: [ImageModelData]
+
+    struct ImageModelData: Codable {
+        let id: String
+        let name: String
+        let description: String?
+        let architecture: Architecture?
+        let supportsStreaming: Bool?
+
+        struct Architecture: Codable {
+            let inputModalities: [String]?
+            let outputModalities: [String]?
+
+            enum CodingKeys: String, CodingKey {
+                case inputModalities = "input_modalities"
+                case outputModalities = "output_modalities"
+            }
+        }
+
+        enum CodingKeys: String, CodingKey {
+            case id, name, description, architecture
+            case supportsStreaming = "supports_streaming"
+        }
+    }
+}
+
+// MARK: - Images API — Generation Response
+
+struct OpenRouterImageGenerationResponse: Codable {
+    let created: Int?
+    let data: [ImageData]
+    let usage: Usage?
+
+    struct ImageData: Codable {
+        let b64Json: String
+        let mediaType: String?
+
+        enum CodingKeys: String, CodingKey {
+            case b64Json = "b64_json"
+            case mediaType = "media_type"
+        }
+    }
+
+    struct Usage: Codable {
+        let promptTokens: Int
+        let completionTokens: Int
+        let totalTokens: Int
+        let cost: Double?
+
+        enum CodingKeys: String, CodingKey {
+            case promptTokens = "prompt_tokens"
+            case completionTokens = "completion_tokens"
+            case totalTokens = "total_tokens"
+            case cost
+        }
+    }
+}
+
 // MARK: - Error Response

 struct OpenRouterErrorResponse: Codable {
@@ -53,30 +53,16 @@ class OpenRouterProvider: AIProvider {
    
    func listModels() async throws -> [ModelInfo] {
        Log.api.info("Fetching model list from OpenRouter")
-        let url = URL(string: "\(baseURL)/models")!
-        var request = URLRequest(url: url)
-        request.addValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
-        request.addValue("application/json", forHTTPHeaderField: "Content-Type")

-        let (data, response) = try await session.data(for: request)
+        // Fetch chat models and image models in parallel
+        async let chatData = fetchRaw(path: "/models")
+        async let imageData = fetchRaw(path: "/images/models")
+        let (chatRaw, imageRaw) = try await (chatData, imageData)

-        guard let httpResponse = response as? HTTPURLResponse else {
-            Log.api.error("OpenRouter models: invalid response (not HTTP)")
-            throw ProviderError.invalidResponse
-        }
+        let modelsResponse = try JSONDecoder().decode(OpenRouterModelsResponse.self, from: chatRaw)
+        Log.api.info("OpenRouter loaded \(modelsResponse.data.count) chat models")

-        guard httpResponse.statusCode == 200 else {
-            if let errorResponse = try? JSONDecoder().decode(OpenRouterErrorResponse.self, from: data) {
-                Log.api.error("OpenRouter models HTTP \(httpResponse.statusCode): \(errorResponse.error.message)")
-                throw ProviderError.unknown(errorResponse.error.message)
-            }
-            Log.api.error("OpenRouter models HTTP \(httpResponse.statusCode)")
-            throw ProviderError.unknown("HTTP \(httpResponse.statusCode)")
-        }
-        
-        let modelsResponse = try JSONDecoder().decode(OpenRouterModelsResponse.self, from: data)
-        Log.api.info("OpenRouter loaded \(modelsResponse.data.count) models")
-        return modelsResponse.data.map { modelData in
+        var models = modelsResponse.data.map { modelData in
            let promptPrice = Double(modelData.pricing.prompt) ?? 0.0
            let completionPrice = Double(modelData.pricing.completion) ?? 0.0
            
@@ -129,8 +115,110 @@ class OpenRouterProvider: AIProvider {
            )
            return info
        }
+
+        // Merge dedicated image models (these don't appear in /models)
+        if let imageModelsResponse = try? JSONDecoder().decode(OpenRouterImageModelsResponse.self, from: imageRaw) {
+            Log.api.info("OpenRouter loaded \(imageModelsResponse.data.count) image models")
+            let existingIds = Set(models.map { $0.id })
+            let imageModels = imageModelsResponse.data.compactMap { m -> ModelInfo? in
+                guard !existingIds.contains(m.id) else { return nil }
+                let acceptsImageInput = m.architecture?.inputModalities?.contains("image") ?? false
+                var info = ModelInfo(
+                    id: m.id,
+                    name: m.name,
+                    description: m.description,
+                    contextLength: 0,
+                    pricing: ModelInfo.Pricing(prompt: 0, completion: 0),
+                    capabilities: ModelInfo.ModelCapabilities(
+                        vision: acceptsImageInput,
+                        tools: false,
+                        online: false,
+                        imageGeneration: true,
+                        thinking: false,
+                        usesImagesAPI: true
+                    ),
+                    topProvider: m.id.components(separatedBy: "/").first
+                )
+                info.categories = ModelCategory.infer(name: m.name, id: m.id, description: m.description)
+                return info
+            }
+            models.append(contentsOf: imageModels)
+        }
+
+        return models
    }
-    
+
+    private func fetchRaw(path: String) async throws -> Data {
+        let url = URL(string: "\(baseURL)\(path)")!
+        var request = URLRequest(url: url)
+        request.addValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
+        request.addValue("application/json", forHTTPHeaderField: "Content-Type")
+        let (data, response) = try await session.data(for: request)
+        guard let httpResponse = response as? HTTPURLResponse else { throw ProviderError.invalidResponse }
+        guard httpResponse.statusCode == 200 else {
+            if let err = try? JSONDecoder().decode(OpenRouterErrorResponse.self, from: data) {
+                throw ProviderError.unknown(err.error.message)
+            }
+            throw ProviderError.unknown("HTTP \(httpResponse.statusCode)")
+        }
+        return data
+    }
+
+    // MARK: - Images API
+
+    func generateImage(model: String, prompt: String) async throws -> ChatResponse {
+        Log.api.info("OpenRouter images API: model=\(model)")
+        let url = URL(string: "\(baseURL)/images")!
+        var urlRequest = URLRequest(url: url)
+        urlRequest.httpMethod = "POST"
+        urlRequest.addValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
+        urlRequest.addValue("application/json", forHTTPHeaderField: "Content-Type")
+        urlRequest.addValue("https://github.com/yourusername/oAI", forHTTPHeaderField: "HTTP-Referer")
+        urlRequest.addValue("oAI-Swift", forHTTPHeaderField: "X-Title")
+        urlRequest.httpBody = try JSONSerialization.data(withJSONObject: ["model": model, "prompt": prompt])
+
+        let (data, response) = try await session.data(for: urlRequest)
+        guard let httpResponse = response as? HTTPURLResponse else { throw ProviderError.invalidResponse }
+
+        if httpResponse.statusCode != 200 {
+            if let err = try? JSONDecoder().decode(OpenRouterErrorResponse.self, from: data) {
+                Log.api.error("OpenRouter images HTTP \(httpResponse.statusCode): \(err.error.message)")
+                throw ProviderError.unknown(err.error.message)
+            }
+            Log.api.error("OpenRouter images HTTP \(httpResponse.statusCode)")
+            throw ProviderError.unknown("HTTP \(httpResponse.statusCode)")
+        }
+
+        if let rawStr = String(data: data, encoding: .utf8) {
+            Log.api.debug("Images API raw response (first 200 chars): \(rawStr.prefix(200))")
+        }
+
+        let imageResponse = try JSONDecoder().decode(OpenRouterImageGenerationResponse.self, from: data)
+        let images: [Data] = imageResponse.data.compactMap { item in
+            Data(base64Encoded: item.b64Json)
+        }
+
+        let usage: ChatResponse.Usage? = imageResponse.usage.map { u in
+            ChatResponse.Usage(
+                promptTokens: u.promptTokens,
+                completionTokens: u.completionTokens,
+                totalTokens: u.totalTokens,
+                rawCostUSD: u.cost
+            )
+        }
+
+        return ChatResponse(
+            id: UUID().uuidString,
+            model: model,
+            content: "",
+            role: "assistant",
+            finishReason: "stop",
+            usage: usage,
+            created: Date(),
+            generatedImages: images.isEmpty ? nil : images
+        )
+    }
+
    func getModel(_ id: String) async throws -> ModelInfo? {
        let models = try await listModels()
        return models.first { $0.id == id }
@@ -801,6 +801,13 @@ Don't narrate future actions ("Let me...") - just use the tools.
        let bashActive = settings.bashEnabled
        let personalDataActive = settings.calendarEnabled || settings.remindersEnabled || settings.contactsEnabled || settings.locationMapsEnabled
        let researchAgentsActive = settings.agentsEnabled
+        // Dedicated images API path (OpenRouter /images endpoint — separate from chat completions)
+        if selectedModel?.capabilities.usesImagesAPI == true,
+           let orProvider = provider as? OpenRouterProvider {
+            generateImageAPIResponse(orProvider: orProvider, modelId: modelId, prompt: prompt)
+            return
+        }
+
        let modelSupportTools = selectedModel?.capabilities.tools ?? false
        if modelSupportTools && (anytypeActive || bashActive || personalDataActive || researchAgentsActive || (mcpActive && !mcp.allowedFolders.isEmpty)) {
            generateAIResponseWithTools(provider: provider, modelId: modelId)
@@ -1263,6 +1270,55 @@ Don't narrate future actions ("Let me...") - just use the tools.

    // MARK: - AI Response with Tool Calls

+    // MARK: - Images API Generation
+
+    private func generateImageAPIResponse(orProvider: OpenRouterProvider, modelId: String, prompt: String) {
+        isGenerating = true
+        streamingTask?.cancel()
+        streamingTask = Task {
+            let startTime = Date()
+            let assistantMessage = Message(
+                role: .assistant,
+                content: ThinkingVerbs.random(),
+                tokens: nil,
+                cost: nil,
+                timestamp: Date(),
+                attachments: nil,
+                modelId: modelId,
+                isStreaming: true
+            )
+            let messageId = assistantMessage.id
+            messages.append(assistantMessage)
+
+            do {
+                let response = try await orProvider.generateImage(model: modelId, prompt: prompt)
+                let responseTime = Date().timeIntervalSince(startTime)
+
+                if let index = messages.firstIndex(where: { $0.id == messageId }) {
+                    messages[index].content = response.content
+                    messages[index].isStreaming = false
+                    messages[index].generatedImages = response.generatedImages
+                    messages[index].responseTime = responseTime
+
+                    if let usage = response.usage {
+                        messages[index].tokens = usage.completionTokens
+                        let cost = usage.rawCostUSD
+                        messages[index].cost = cost
+                        sessionStats.addMessage(inputTokens: usage.promptTokens, outputTokens: usage.completionTokens, cost: cost)
+                    }
+                }
+                _ = detectGoodbyePhrase(in: "")
+            } catch {
+                if let index = messages.firstIndex(where: { $0.id == messageId }) {
+                    messages[index].content = "❌ Image generation failed: \(error.localizedDescription)"
+                    messages[index].isStreaming = false
+                }
+                Log.api.error("Images API error: \(error)")
+            }
+            isGenerating = false
+        }
+    }
+
    private func generateAIResponseWithTools(provider: AIProvider, modelId: String) {
        let mcp = MCPService.shared
        Log.ui.info("generateAIResponseWithTools: model=\(modelId)")