omnikey-cli 1.1.0 → 1.3.0

package/backend-dist/__tests__/ai-client.adapters.test.js

@@ -0,0 +1,247 @@
+"use strict";
+/**
+ * Per-provider adapter tests for temperature handling.
+ *
+ * The three SDKs (`openai`, `@anthropic-ai/sdk`, `@google/genai`) are mocked
+ * at the module boundary using `vi.mock`. Mock spies are declared inside a
+ * `vi.hoisted()` block so they are available when `vi.mock` factories run
+ * (vi.mock is hoisted to the top of the file).
+ *
+ * These tests never contact any real API.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+const vitest_1 = require("vitest");
+const mocks = vitest_1.vi.hoisted(() => ({
+    openaiCreate: vitest_1.vi.fn(),
+    anthropicCreate: vitest_1.vi.fn(),
+    anthropicStream: vitest_1.vi.fn(),
+    geminiGenerate: vitest_1.vi.fn(),
+    geminiGenerateStream: vitest_1.vi.fn(),
+}));
+vitest_1.vi.mock('openai', () => ({
+    default: class MockOpenAI {
+        constructor(_opts) {
+            this.chat = { completions: { create: mocks.openaiCreate } };
+            this.images = { generate: vitest_1.vi.fn() };
+        }
+    },
+}));
+vitest_1.vi.mock('@anthropic-ai/sdk', () => ({
+    default: class MockAnthropic {
+        constructor(_opts) {
+            this.messages = { create: mocks.anthropicCreate, stream: mocks.anthropicStream };
+        }
+    },
+}));
+vitest_1.vi.mock('@google/genai', () => ({
+    GoogleGenAI: class MockGoogleGenAI {
+        constructor(_opts) {
+            this.models = {
+                generateContent: mocks.geminiGenerate,
+                generateContentStream: mocks.geminiGenerateStream,
+                generateImages: vitest_1.vi.fn(),
+            };
+        }
+    },
+    // The adapter file imports these as types-only but they still need to resolve.
+    Content: class {
+    },
+    Tool: class {
+    },
+}));
+const ai_client_1 = require("../ai-client");
+const messages = [{ role: 'user', content: 'hello' }];
+function asAsyncIterable(chunks) {
+    return {
+        [Symbol.asyncIterator]: async function* () {
+            for (const c of chunks)
+                yield c;
+        },
+    };
+}
+(0, vitest_1.beforeEach)(() => {
+    mocks.openaiCreate.mockReset();
+    mocks.anthropicCreate.mockReset();
+    mocks.anthropicStream.mockReset();
+    mocks.geminiGenerate.mockReset();
+    mocks.geminiGenerateStream.mockReset();
+});
+// ---------------------------------------------------------------------------
+// OpenAI
+// ---------------------------------------------------------------------------
+(0, vitest_1.describe)('OpenAIAdapter temperature handling', () => {
+    function mockCompleteResponse() {
+        mocks.openaiCreate.mockResolvedValueOnce({
+            choices: [{ message: { content: 'ok', tool_calls: undefined }, finish_reason: 'stop' }],
+            usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 },
+        });
+    }
+    function mockStreamResponse() {
+        mocks.openaiCreate.mockResolvedValueOnce(asAsyncIterable([
+            { choices: [{ delta: { content: 'ok' } }] },
+            {
+                choices: [{ delta: {} }],
+                usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 },
+            },
+        ]));
+    }
+    (0, vitest_1.it)('complete: passes temperature for gpt-4o-mini', async () => {
+        mockCompleteResponse();
+        const client = new ai_client_1.AIClient('openai', 'sk-test');
+        await client.complete('gpt-4o-mini', messages, { temperature: 0.42 });
+        const body = mocks.openaiCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).toHaveProperty('temperature', 0.42);
+    });
+    (0, vitest_1.it)('complete: omits temperature for gpt-5.5 even if caller supplies one', async () => {
+        mockCompleteResponse();
+        const client = new ai_client_1.AIClient('openai', 'sk-test');
+        await client.complete('gpt-5.5', messages, { temperature: 0.42 });
+        const body = mocks.openaiCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).not.toHaveProperty('temperature');
+    });
+    vitest_1.it.each(['gpt-5', 'gpt-5-mini', 'gpt-5.1', 'o1', 'o3-mini', 'o4-mini'])('complete: omits temperature for unsupported model %s', async (model) => {
+        mockCompleteResponse();
+        const client = new ai_client_1.AIClient('openai', 'sk-test');
+        await client.complete(model, messages, { temperature: 0.7 });
+        const body = mocks.openaiCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).not.toHaveProperty('temperature');
+    });
+    (0, vitest_1.it)('streamComplete: passes temperature for gpt-4o-mini', async () => {
+        mockStreamResponse();
+        const client = new ai_client_1.AIClient('openai', 'sk-test');
+        await client.streamComplete('gpt-4o-mini', messages, { temperature: 0.31 }, () => { });
+        const body = mocks.openaiCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).toHaveProperty('temperature', 0.31);
+        (0, vitest_1.expect)(body).toHaveProperty('stream', true);
+    });
+    (0, vitest_1.it)('streamComplete: omits temperature for gpt-5.5', async () => {
+        mockStreamResponse();
+        const client = new ai_client_1.AIClient('openai', 'sk-test');
+        await client.streamComplete('gpt-5.5', messages, { temperature: 0.31 }, () => { });
+        const body = mocks.openaiCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).not.toHaveProperty('temperature');
+        (0, vitest_1.expect)(body).toHaveProperty('stream', true);
+    });
+    (0, vitest_1.it)('streamComplete: omits temperature even when caller passes empty options for gpt-5.5', async () => {
+        mockStreamResponse();
+        const client = new ai_client_1.AIClient('openai', 'sk-test');
+        await client.streamComplete('gpt-5.5', messages, {}, () => { });
+        const body = mocks.openaiCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).not.toHaveProperty('temperature');
+    });
+    (0, vitest_1.it)('streamComplete: uses 0.3 default for supported model when caller omits temperature', async () => {
+        mockStreamResponse();
+        const client = new ai_client_1.AIClient('openai', 'sk-test');
+        await client.streamComplete('gpt-4o-mini', messages, {}, () => { });
+        const body = mocks.openaiCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).toHaveProperty('temperature', 0.3);
+    });
+});
+// ---------------------------------------------------------------------------
+// Anthropic
+// ---------------------------------------------------------------------------
+(0, vitest_1.describe)('AnthropicAdapter temperature handling', () => {
+    function mockCompleteResponse() {
+        mocks.anthropicCreate.mockResolvedValueOnce({
+            content: [{ type: 'text', text: 'ok' }],
+            stop_reason: 'end_turn',
+            usage: { input_tokens: 1, output_tokens: 1 },
+        });
+    }
+    function mockStreamResponse() {
+        const finalMessage = vitest_1.vi.fn().mockResolvedValue({
+            usage: { input_tokens: 1, output_tokens: 1 },
+        });
+        const stream = asAsyncIterable([
+            { type: 'content_block_delta', delta: { type: 'text_delta', text: 'ok' } },
+        ]);
+        stream.finalMessage = finalMessage;
+        mocks.anthropicStream.mockReturnValueOnce(stream);
+    }
+    (0, vitest_1.it)('complete: passes temperature for claude-sonnet-4-5', async () => {
+        mockCompleteResponse();
+        const client = new ai_client_1.AIClient('anthropic', 'sk-anthropic-test');
+        await client.complete('claude-sonnet-4-5', messages, { temperature: 0.42 });
+        const body = mocks.anthropicCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).toHaveProperty('temperature', 0.42);
+    });
+    vitest_1.it.each([
+        'claude-haiku-4-5-20251001',
+        'claude-opus-4-5',
+        'claude-opus-4-5-20251101',
+        'claude-opus-4-6',
+    ])('complete: passes temperature for supported model %s', async (model) => {
+        mockCompleteResponse();
+        const client = new ai_client_1.AIClient('anthropic', 'sk-anthropic-test');
+        await client.complete(model, messages, { temperature: 0.5 });
+        const body = mocks.anthropicCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).toHaveProperty('temperature', 0.5);
+    });
+    vitest_1.it.each(['claude-opus-4-7', 'claude-opus-4-7-20260101'])('complete: omits temperature for unsupported model %s', async (model) => {
+        mockCompleteResponse();
+        const client = new ai_client_1.AIClient('anthropic', 'sk-anthropic-test');
+        await client.complete(model, messages, { temperature: 0.5 });
+        const body = mocks.anthropicCreate.mock.calls[0][0];
+        (0, vitest_1.expect)(body).not.toHaveProperty('temperature');
+    });
+    (0, vitest_1.it)('streamComplete: passes temperature for claude-sonnet-4-5', async () => {
+        mockStreamResponse();
+        const client = new ai_client_1.AIClient('anthropic', 'sk-anthropic-test');
+        await client.streamComplete('claude-sonnet-4-5', messages, { temperature: 0.6 }, () => { });
+        const body = mocks.anthropicStream.mock.calls[0][0];
+        (0, vitest_1.expect)(body).toHaveProperty('temperature', 0.6);
+    });
+    (0, vitest_1.it)('streamComplete: omits temperature for claude-opus-4-7', async () => {
+        mockStreamResponse();
+        const client = new ai_client_1.AIClient('anthropic', 'sk-anthropic-test');
+        await client.streamComplete('claude-opus-4-7', messages, { temperature: 0.6 }, () => { });
+        const body = mocks.anthropicStream.mock.calls[0][0];
+        (0, vitest_1.expect)(body).not.toHaveProperty('temperature');
+    });
+});
+// ---------------------------------------------------------------------------
+// Gemini
+// ---------------------------------------------------------------------------
+(0, vitest_1.describe)('GeminiAdapter temperature handling', () => {
+    function mockCompleteResponse() {
+        mocks.geminiGenerate.mockResolvedValueOnce({
+            candidates: [
+                {
+                    content: { parts: [{ text: 'ok' }] },
+                    finishReason: 'STOP',
+                },
+            ],
+            usageMetadata: { promptTokenCount: 1, candidatesTokenCount: 1, totalTokenCount: 2 },
+        });
+    }
+    function mockStreamResponse() {
+        mocks.geminiGenerateStream.mockResolvedValueOnce(asAsyncIterable([
+            { text: 'ok' },
+            {
+                text: '',
+                usageMetadata: { promptTokenCount: 1, candidatesTokenCount: 1, totalTokenCount: 2 },
+            },
+        ]));
+    }
+    vitest_1.it.each(['gemini-2.5-flash', 'gemini-2.5-pro', 'gemini-3-pro'])('complete: passes temperature for %s (all Gemini models accept it)', async (model) => {
+        mockCompleteResponse();
+        const client = new ai_client_1.AIClient('gemini', 'gemini-test-key');
+        await client.complete(model, messages, { temperature: 0.42 });
+        const body = mocks.geminiGenerate.mock.calls[0][0];
+        (0, vitest_1.expect)(body.config).toHaveProperty('temperature', 0.42);
+    });
+    (0, vitest_1.it)('streamComplete: passes temperature for gemini-2.5-pro', async () => {
+        mockStreamResponse();
+        const client = new ai_client_1.AIClient('gemini', 'gemini-test-key');
+        await client.streamComplete('gemini-2.5-pro', messages, { temperature: 0.31 }, () => { });
+        const body = mocks.geminiGenerateStream.mock.calls[0][0];
+        (0, vitest_1.expect)(body.config).toHaveProperty('temperature', 0.31);
+    });
+    (0, vitest_1.it)('streamComplete: applies default 0.3 when caller omits temperature', async () => {
+        mockStreamResponse();
+        const client = new ai_client_1.AIClient('gemini', 'gemini-test-key');
+        await client.streamComplete('gemini-2.5-pro', messages, {}, () => { });
+        const body = mocks.geminiGenerateStream.mock.calls[0][0];
+        (0, vitest_1.expect)(body.config).toHaveProperty('temperature', 0.3);
+    });
+});

package/backend-dist/__tests__/ai-client.helpers.test.js

@@ -0,0 +1,99 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+const vitest_1 = require("vitest");
+const ai_client_1 = require("../ai-client");
+(0, vitest_1.describe)('modelSupportsTemperature', () => {
+    (0, vitest_1.describe)('OpenAI', () => {
+        vitest_1.it.each([
+            ['gpt-4o-mini', true],
+            ['gpt-4o', true],
+            ['gpt-4-turbo', true],
+            ['gpt-3.5-turbo', true],
+        ])('allows temperature for %s', (model, expected) => {
+            (0, vitest_1.expect)((0, ai_client_1.modelSupportsTemperature)(model)).toBe(expected);
+        });
+        vitest_1.it.each([
+            ['gpt-5', false],
+            ['gpt-5-mini', false],
+            ['gpt-5.1', false],
+            ['gpt-5.5', false],
+            ['GPT-5.5', false], // case-insensitive
+        ])('rejects temperature for GPT-5 family member %s', (model, expected) => {
+            (0, vitest_1.expect)((0, ai_client_1.modelSupportsTemperature)(model)).toBe(expected);
+        });
+        vitest_1.it.each([
+            ['o1', false],
+            ['o1-preview', false],
+            ['o3', false],
+            ['o3-mini', false],
+            ['o4-mini', false],
+        ])('rejects temperature for reasoning model %s', (model, expected) => {
+            (0, vitest_1.expect)((0, ai_client_1.modelSupportsTemperature)(model)).toBe(expected);
+        });
+    });
+    (0, vitest_1.describe)('Gemini', () => {
+        vitest_1.it.each([
+            ['gemini-2.5-flash', true],
+            ['gemini-2.5-pro', true],
+            ['gemini-3-pro', true],
+            ['gemini-3.5-flash', true],
+        ])('allows temperature for %s', (model, expected) => {
+            (0, vitest_1.expect)((0, ai_client_1.modelSupportsTemperature)(model)).toBe(expected);
+        });
+    });
+    (0, vitest_1.describe)('Anthropic', () => {
+        vitest_1.it.each([
+            ['claude-haiku-4-5', true],
+            ['claude-haiku-4-5-20251001', true],
+            ['claude-sonnet-4-5', true],
+            ['claude-sonnet-4-5-20250929', true],
+            ['claude-opus-4-5', true],
+            ['claude-opus-4-5-20251101', true],
+            ['claude-opus-4-6', true],
+        ])('allows temperature for %s', (model, expected) => {
+            (0, vitest_1.expect)((0, ai_client_1.modelSupportsTemperature)(model)).toBe(expected);
+        });
+        vitest_1.it.each([
+            ['claude-opus-4-7', false],
+            ['claude-opus-4-7-20260101', false],
+            ['CLAUDE-OPUS-4-7', false], // case-insensitive
+        ])('rejects temperature for opus-4-7 variant %s', (model, expected) => {
+            (0, vitest_1.expect)((0, ai_client_1.modelSupportsTemperature)(model)).toBe(expected);
+        });
+    });
+});
+(0, vitest_1.describe)('getDefaultModel', () => {
+    (0, vitest_1.it)('returns the configured fast and smart tiers for each provider', () => {
+        // Don't pin exact model strings — they will be upgraded over time. Just
+        // assert that each provider returns a non-empty string for both tiers
+        // and that fast/smart differ (smart is meant to be a bigger model).
+        for (const provider of ['openai', 'gemini', 'anthropic']) {
+            const fast = (0, ai_client_1.getDefaultModel)(provider, 'fast');
+            const smart = (0, ai_client_1.getDefaultModel)(provider, 'smart');
+            (0, vitest_1.expect)(fast).toBeTruthy();
+            (0, vitest_1.expect)(smart).toBeTruthy();
+            (0, vitest_1.expect)(fast).not.toEqual(smart);
+        }
+    });
+    (0, vitest_1.it)('returns smart-tier models that are correctly classified by modelSupportsTemperature', () => {
+        // Regression guard: whenever a smart model is upgraded, the helper must
+        // continue to return the correct policy for it. This test is the single
+        // place that ties the two together so an accidental mismatch breaks the
+        // suite immediately.
+        const expectations = {
+            // OpenAI smart tier is in the GPT-5 family → no temperature.
+            openai: false,
+            // Gemini smart tier accepts temperature.
+            gemini: true,
+            // Anthropic smart tier is claude-opus-4-7 → no temperature.
+            anthropic: false,
+        };
+        for (const provider of Object.keys(expectations)) {
+            const expected = expectations[provider];
+            if (expected === null)
+                continue;
+            const smartModel = (0, ai_client_1.getDefaultModel)(provider, 'smart');
+            (0, vitest_1.expect)((0, ai_client_1.modelSupportsTemperature)(smartModel), `${provider} smart model "${smartModel}" should report temperature-support=${expected}`).toBe(expected);
+        }
+    });
+});

package/backend-dist/__tests__/featureRoutes.runEnhancementModel.test.js

@@ -0,0 +1,79 @@
+"use strict";
+/**
+ * Tests for the temperature-handling change in `runEnhancementModel`.
+ *
+ * - 'enhance' → { temperature: 0.3 }
+ * - 'grammar' → { temperature: 0.3 }
+ * - 'task'    → {}   (no temperature; smart-tier model decides for itself)
+ *
+ * Mocks `./ai-client` and `./models/subscriptionTaskTemplate` so the test
+ * stays a pure unit test and never touches the database or any SDK.
+ */
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+const vitest_1 = require("vitest");
+const winston_1 = __importDefault(require("winston"));
+const mocks = vitest_1.vi.hoisted(() => ({
+    streamComplete: vitest_1.vi.fn(),
+    getDefaultModel: vitest_1.vi.fn(),
+    findOne: vitest_1.vi.fn(),
+}));
+vitest_1.vi.mock('../ai-client', () => ({
+    aiClient: { streamComplete: mocks.streamComplete },
+    getDefaultModel: mocks.getDefaultModel,
+}));
+vitest_1.vi.mock('../models/subscriptionTaskTemplate', () => ({
+    SubscriptionTaskTemplate: { findOne: mocks.findOne },
+}));
+const featureRoutes_1 = require("../featureRoutes");
+function makeLogger() {
+    return winston_1.default.createLogger({
+        silent: true,
+        transports: [new winston_1.default.transports.Console({ silent: true })],
+    });
+}
+const fakeSubscription = { id: 'sub_test' };
+(0, vitest_1.beforeEach)(() => {
+    mocks.streamComplete.mockReset();
+    mocks.streamComplete.mockResolvedValue({ usage: undefined, model: 'mock-model' });
+    mocks.getDefaultModel.mockReset();
+    mocks.getDefaultModel.mockImplementation((_provider, tier) => tier === 'smart' ? 'smart-model-mock' : 'fast-model-mock');
+    mocks.findOne.mockReset();
+    // Default task template — plain text passes through `decompressString`
+    // so `getPromptForCommand('task', ...)` returns a non-empty prompt and the
+    // streamComplete path is reached.
+    mocks.findOne.mockResolvedValue({ instructions: 'You are a helpful task assistant.' });
+});
+(0, vitest_1.describe)('runEnhancementModel — temperature per command', () => {
+    (0, vitest_1.it)("passes temperature: 0.3 for cmd='enhance'", async () => {
+        const result = await (0, featureRoutes_1.runEnhancementModel)(makeLogger(), 'hello world', 'enhance', fakeSubscription);
+        (0, vitest_1.expect)(result).not.toBeNull();
+        (0, vitest_1.expect)(mocks.streamComplete).toHaveBeenCalledTimes(1);
+        const [, , options] = mocks.streamComplete.mock.calls[0];
+        (0, vitest_1.expect)(options).toEqual({ temperature: 0.3 });
+    });
+    (0, vitest_1.it)("passes temperature: 0.3 for cmd='grammar'", async () => {
+        const result = await (0, featureRoutes_1.runEnhancementModel)(makeLogger(), 'helo wrld', 'grammar', fakeSubscription);
+        (0, vitest_1.expect)(result).not.toBeNull();
+        (0, vitest_1.expect)(mocks.streamComplete).toHaveBeenCalledTimes(1);
+        const [, , options] = mocks.streamComplete.mock.calls[0];
+        (0, vitest_1.expect)(options).toEqual({ temperature: 0.3 });
+    });
+    (0, vitest_1.it)("omits temperature for cmd='task' (custom-task)", async () => {
+        const result = await (0, featureRoutes_1.runEnhancementModel)(makeLogger(), 'do the thing', 'task', fakeSubscription);
+        (0, vitest_1.expect)(result).not.toBeNull();
+        (0, vitest_1.expect)(mocks.streamComplete).toHaveBeenCalledTimes(1);
+        const [, , options] = mocks.streamComplete.mock.calls[0];
+        (0, vitest_1.expect)(options).toEqual({});
+        (0, vitest_1.expect)(options).not.toHaveProperty('temperature');
+    });
+    (0, vitest_1.it)("selects the smart-tier model for cmd='task' and fast-tier for enhance/grammar", async () => {
+        await (0, featureRoutes_1.runEnhancementModel)(makeLogger(), 'a', 'task', fakeSubscription);
+        await (0, featureRoutes_1.runEnhancementModel)(makeLogger(), 'b', 'enhance', fakeSubscription);
+        await (0, featureRoutes_1.runEnhancementModel)(makeLogger(), 'c', 'grammar', fakeSubscription);
+        const modelsCalled = mocks.streamComplete.mock.calls.map(([model]) => model);
+        (0, vitest_1.expect)(modelsCalled).toEqual(['smart-model-mock', 'fast-model-mock', 'fast-model-mock']);
+    });
+});

package/backend-dist/agent/agentServer.js

@@ -603,19 +603,19 @@ async function runAgentTurnInternal(sessionId, subscription, clientMessage, send
             // is not the final turn (e.g. plain-text conclusion after terminal
             // output). Treat it as a final answer so the client is never left
             // hanging.
-            log.info('Agent returned untagged content on a non-final turn; treating as final answer', {
+            log.info('Agent returned untagged content on a non-final turn; treating as assistant response and looping the function again.', {
                 sessionId,
                 subscriptionId: subscription.id,
                 turn: session.turns,
             });
             (0, utils_1.pushToSessionHistory)(log, session, { role: 'assistant', content });
             await persistSessionToDB(sessionId, session);
-            send({
-                session_id: sessionId,
+            await runAgentTurnInternal(sessionId, subscription, {
                 sender: 'agent',
-                content: `<final_answer>\n${content}\n</final_answer>`,
-            });
-            void (0, sessionGrouping_1.updateSessionGroup)(sessionId, subscription.id);
+                session_id: sessionId,
+                content: '',
+                is_web_call: true,
+            }, send, logger_1.logger, options);
         }
         else {
             log.warn('Agent returned empty content with no recognized tags; sending error', {
@@ -797,7 +797,12 @@ function buildTranscript(raw) {
                 break;
             }
         }
-        currentAssistant.text = finalText || blocks.map((b) => b.text).join('\n\n').trim();
+        currentAssistant.text =
+            finalText ||
+                blocks
+                    .map((b) => b.text)
+                    .join('\n\n')
+                    .trim();
         messages.push(currentAssistant);
         currentAssistant = null;
     };

package/backend-dist/ai-client.js

@@ -5,6 +5,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.aiClient = exports.AIClient = void 0;
 exports.getDefaultModel = getDefaultModel;
+exports.modelSupportsTemperature = modelSupportsTemperature;
 exports.getMaxMessageContentLength = getMaxMessageContentLength;
 exports.getMaxHistoryLength = getMaxHistoryLength;
 exports.getContextWindowSize = getContextWindowSize;
@@ -17,6 +18,12 @@ const config_1 = require("./config");
 // Default model mapping
 // ---------------------------------------------------------------------------
 const DEFAULT_MODELS = {
+    // Smart-tier picks track each provider's current flagship for
+    // reasoning/coding workloads. Update here when a newer model becomes
+    // generally available so both the feature routes and the agent server pick
+    // it up automatically. When swapping a smart model in, also verify whether
+    // it accepts the `temperature` parameter and update
+    // `modelSupportsTemperature` accordingly.
     openai: { fast: 'gpt-4o-mini', smart: 'gpt-5.5' },
     gemini: { fast: 'gemini-2.5-flash', smart: 'gemini-2.5-pro' },
     anthropic: { fast: 'claude-haiku-4-5-20251001', smart: 'claude-opus-4-7' },
@@ -24,6 +31,39 @@ const DEFAULT_MODELS = {
 function getDefaultModel(provider, tier) {
     return DEFAULT_MODELS[provider][tier];
 }
+/**
+ * Returns whether a given model accepts the `temperature` parameter.
+ *
+ * Provider-specific rules (validated against published API docs and SDKs as
+ * of late 2025 / early 2026):
+ *  - OpenAI GPT-5 family (`gpt-5`, `gpt-5-mini`, `gpt-5.1`, …): NOT supported.
+ *    The API only accepts the default value (1) and returns
+ *    `unsupported_value: 'temperature'` for anything else.
+ *  - OpenAI o-series reasoning models (`o1`, `o3`, `o4-mini`, …): NOT
+ *    supported for the same reason.
+ *  - OpenAI GPT-4 / GPT-4o / GPT-3.5: supported.
+ *  - Google Gemini (2.x and 3.x families): supported via `generationConfig`.
+ *  - Anthropic Claude (Sonnet, Haiku, and Opus 4.x): supported, with the
+ *    exception of `claude-opus-4-7` (and its dated revisions) which rejects
+ *    `temperature` just like the OpenAI GPT-5 family.
+ */
+function modelSupportsTemperature(model) {
+    // OpenAI GPT-5 family (gpt-5, gpt-5-mini, gpt-5.1, gpt-5.5, …) only
+    // accepts the default temperature (1) — anything else is rejected with
+    // `unsupported_value: 'temperature'`.
+    if (/^gpt-5(\b|[.\-])/i.test(model))
+        return false;
+    // OpenAI o-series reasoning models (o1, o3, o4-mini, …) likewise drop the
+    // `temperature` knob.
+    if (/^o[134](\b|[-_])/i.test(model))
+        return false;
+    // Anthropic's Claude Opus 4.7 line (and its dated revisions like
+    // `claude-opus-4-7-20260101`) does not accept `temperature`; the rest of
+    // the Claude 4.x family (Sonnet, Haiku, Opus 4.5/4.6) does.
+    if (/^claude-opus-4-7(\b|[-_])/i.test(model))
+        return false;
+    return true;
+}
 /**
  * Maximum character length for a single message content string per provider.
  *
@@ -88,7 +128,7 @@ class OpenAIAdapter {
             model,
             messages: oaiMessages,
             tools: tools?.length ? tools : undefined,
-            temperature: model === 'gpt-5.5' ? 1 : (options.temperature ?? 0.2),
+            ...(modelSupportsTemperature(model) ? { temperature: options.temperature ?? 0.2 } : {}),
             max_tokens: options.maxTokens,
         });
         const choice = completion.choices[0];
@@ -125,7 +165,9 @@ class OpenAIAdapter {
         const stream = await this.client.chat.completions.create({
             model,
             messages: oaiMessages,
-            temperature: options.temperature ?? 0.3,
+            ...(modelSupportsTemperature(model)
+                ? { temperature: options.temperature ?? 0.3 }
+                : {}),
             stream: true,
             stream_options: { include_usage: true },
         });
@@ -188,7 +230,7 @@ class AnthropicAdapter {
             ...(system ? { system } : {}),
             messages: anthropicMessages,
             ...(tools?.length ? { tools } : {}),
-            ...(model === 'claude-opus-4-7' ? {} : { temperature: options.temperature ?? 0.2 }),
+            ...(modelSupportsTemperature(model) ? { temperature: options.temperature ?? 0.2 } : {}),
         });
         const textContent = response.content
             .filter((b) => b.type === 'text')
@@ -235,7 +277,9 @@ class AnthropicAdapter {
             max_tokens: options.maxTokens ?? 8192,
             ...(system ? { system } : {}),
             messages: anthropicMessages,
-            temperature: options.temperature ?? 0.3,
+            ...(modelSupportsTemperature(model)
+                ? { temperature: options.temperature ?? 0.3 }
+                : {}),
         });
         for await (const event of stream) {
             if (event.type === 'content_block_delta' &&
@@ -269,7 +313,9 @@ class GeminiAdapter {
             config: {
                 ...(systemInstruction ? { systemInstruction } : {}),
                 ...(tools?.length ? { tools } : {}),
-                temperature: options.temperature ?? 0.2,
+                ...(modelSupportsTemperature(model)
+                    ? { temperature: options.temperature ?? 0.2 }
+                    : {}),
             },
         });
         const candidate = response.candidates?.[0];
@@ -320,7 +366,9 @@ class GeminiAdapter {
             contents,
             config: {
                 ...(systemInstruction ? { systemInstruction } : {}),
-                temperature: options.temperature ?? 0.3,
+                ...(modelSupportsTemperature(model)
+                    ? { temperature: options.temperature ?? 0.3 }
+                    : {}),
             },
         });
         let usage;

package/backend-dist/db.js

@@ -44,6 +44,68 @@ async function runSQLiteMigrations(logger) {
             logger.info(`SQLite migration: added column ${table}.${column}`);
         }
     }
+    // mcp_servers was originally created with UNIQUE on both subscription_id and
+    // name as column-level constraints (SQLite auto-indexes). These can't be
+    // dropped with DROP INDEX — the only fix is to recreate the table with the
+    // correct schema (composite unique on subscription_id+name only).
+    await migrateMcpServersTableIfNeeded(logger);
+}
+async function migrateMcpServersTableIfNeeded(logger) {
+    // Check if the old schema is still in place by inspecting the CREATE TABLE sql.
+    const rows = (await sequelize.query(`SELECT sql FROM sqlite_master WHERE type='table' AND name='mcp_servers'`))[0];
+    if (!rows.length)
+        return; // table doesn't exist yet — sync() will create it correctly
+    const createSql = rows[0].sql;
+    // Old schema has UNIQUE on subscription_id at the column level.
+    // New schema only has the composite index mcp_servers_subscription_id_name.
+    const needsMigration = /`subscription_id`[^,]*UNIQUE/i.test(createSql);
+    if (!needsMigration)
+        return;
+    logger.info('SQLite migration: recreating mcp_servers table to remove stale UNIQUE constraints');
+    await sequelize.query('PRAGMA foreign_keys = OFF');
+    try {
+        await sequelize.query('BEGIN TRANSACTION');
+        await sequelize.query(`
+      CREATE TABLE \`mcp_servers_new\` (
+        \`id\` VARCHAR(255) NOT NULL PRIMARY KEY,
+        \`subscription_id\` VARCHAR(255) NOT NULL REFERENCES \`subscriptions\` (\`id\`) ON DELETE CASCADE ON UPDATE CASCADE,
+        \`name\` VARCHAR(100) NOT NULL,
+        \`description\` VARCHAR(500),
+        \`transport\` VARCHAR(16) NOT NULL DEFAULT 'stdio',
+        \`command\` VARCHAR(500),
+        \`args\` JSON NOT NULL DEFAULT '[]',
+        \`env\` JSON NOT NULL DEFAULT '{}',
+        \`url\` VARCHAR(1000),
+        \`headers\` JSON NOT NULL DEFAULT '{}',
+        \`is_enabled\` TINYINT(1) NOT NULL DEFAULT 1,
+        \`last_connected_at\` DATETIME,
+        \`last_error\` TEXT,
+        \`createdAt\` DATETIME NOT NULL,
+        \`updatedAt\` DATETIME NOT NULL
+      )
+    `);
+        await sequelize.query(`
+      INSERT INTO \`mcp_servers_new\`
+        SELECT id, subscription_id, name, description, transport, command, args, env,
+               url, headers, is_enabled, last_connected_at, last_error, createdAt, updatedAt
+        FROM \`mcp_servers\`
+    `);
+        await sequelize.query('DROP TABLE `mcp_servers`');
+        await sequelize.query('ALTER TABLE `mcp_servers_new` RENAME TO `mcp_servers`');
+        await sequelize.query(`
+      CREATE UNIQUE INDEX IF NOT EXISTS \`mcp_servers_subscription_id_name\`
+        ON \`mcp_servers\` (\`subscription_id\`, \`name\`)
+    `);
+        await sequelize.query('COMMIT');
+        logger.info('SQLite migration: mcp_servers table recreated successfully');
+    }
+    catch (err) {
+        await sequelize.query('ROLLBACK');
+        throw err;
+    }
+    finally {
+        await sequelize.query('PRAGMA foreign_keys = ON');
+    }
 }
 async function initDatabase(logger) {
     try {

package/backend-dist/featureRoutes.js

@@ -56,13 +56,12 @@ async function getPromptForCommand(logger, cmd, subscription) {
     return '';
 }
 function getModelForCommand(cmd) {
+    // 'task' is the custom-task command and routes to the smart-tier model.
+    // 'enhance' and 'grammar' use the fast tier. The actual model strings live
+    // in ai-client.ts (DEFAULT_MODELS) so all callers stay in sync when we
+    // upgrade to a newer flagship model.
     const tier = cmd === 'task' ? 'smart' : 'fast';
-    const models = {
-        openai: { fast: 'gpt-4o-mini', smart: 'gpt-5.5' },
-        gemini: { fast: 'gemini-2.5-flash', smart: 'gemini-2.5-pro' },
-        anthropic: { fast: 'claude-haiku-4-5-20251001', smart: 'claude-opus-4-7' },
-    };
-    return models[config_1.config.aiProvider]?.[tier] ?? 'gpt-4o-mini';
+    return (0, ai_client_1.getDefaultModel)(config_1.config.aiProvider, tier);
 }
 function createMessagesParams(cmd, input, prompt) {
     if (cmd === 'task') {
@@ -100,7 +99,14 @@ async function runEnhancementModel(logger, text, cmd, subscription, onDelta) {
     const messages = createMessagesParams(cmd, trimmed, prompt);
     let rawResponse = '';
     let usage;
-    const result = await ai_client_1.aiClient.streamComplete(model, messages, { temperature: 0.3 }, (delta) => {
+    // Smart-tier models (used by the custom-task command) include OpenAI's
+    // GPT-5 family, which rejects any non-default `temperature`. Even on
+    // providers where the smart model still accepts it (Gemini, Anthropic),
+    // omitting `temperature` keeps the request shape uniform across providers
+    // and lets each model use its own tuned default. The fast-tier models used
+    // by `enhance` and `grammar` keep the previous 0.3 default.
+    const completionOptions = cmd === 'task' ? {} : { temperature: 0.3 };
+    const result = await ai_client_1.aiClient.streamComplete(model, messages, completionOptions, (delta) => {
         rawResponse += delta;
         if (onDelta)
             onDelta(delta);

package/backend-dist/index.js

@@ -107,7 +107,7 @@ app.get('/macos/appcast', (req, res) => {
 // ── Windows distribution endpoints ───────────────────────────────────────────
 // These should match the values in windows/OmniKey.Windows.csproj
 // <Version> and windows/build_release_zip.ps1 $APP_VERSION.
-const WIN_VERSION = '1.12';
+const WIN_VERSION = '1.13';
 const WIN_ZIP_FILENAME = 'OmniKeyAI-windows-win-x64.zip';
 const WIN_ZIP_PATH = path_1.default.join(process.cwd(), 'windows', WIN_ZIP_FILENAME);
 // Serves the pre-built ZIP produced by windows/build_release_zip.ps1.
@@ -152,15 +152,19 @@ app.get('/windows/update', (req, res) => {
         releaseNotes: [
             `What's new in ${WIN_VERSION}`,
             ``,
-            `• Brand-new WPF shell — Agent Chat, OmniAgent Session, Task Instructions, Scheduled Jobs, Job Run History, MCP Servers, Manual, Subscription, and Check Updates all reachable from a unified sidebar.`,
-            `• Agent Chat with session sidebar, search, hover-to-delete, context-window indicator, default task-instruction picker (with a built-in "No task instructions" option).`,
-            `• OmniAgent Session replaces the old thinking window — sticky session picker, streaming collapsible timeline (web search, MCP, reasoning, terminal) and a copyable final-answer card. Selecting an existing session previews its last turn in-place.`,
-            `• Scheduled Jobs redesigned: clearer labels, self-contained Active toggle, restructured Save / Run now / Reset / Delete action bar, tooltips throughout.`,
-            `• Theme reworked to match the app icon — cyan + light purple. The Windows system-accent leak (orange/red buttons) is gone everywhere.`,
-            `• Modern typography (Aptos / Inter / Segoe UI Variable Display).`,
-            `• Mouse-wheel scroll fixed across Chat and OmniAgent pages.`,
-            `• Production-default backend URL: shipping binary defaults to https://omnikeyai.ca.`,
-            `• Internals: removed ~5,200 lines of unwired WinForms code; build is warning-free.`,
+            `• Projects: chats are now grouped by project in the sidebar — collapsible "folder" headers per group, a session count badge, and per-header collapse state that survives streaming turns.`,
+            `• Projects: new project picker in the composer toolbar (next to the task-instruction selector) — pick the project for your next turn, mirrors the macOS "Select project" menu. Auto-hides until the backend has classified at least one group.`,
+            `• Projects: the chosen project is stamped onto the outbound message and the optimistic session placeholder, so new chats appear under the right header immediately.`,
+            `• Chat: messages now sit in a centered 820 DIP reading column on large monitors and stretch edge-to-edge on smaller windows — matches the macOS layout exactly. User bubbles stay pinned right, assistant content stays pinned left, on every viewport.`,
+            `• Chat: new Final Answer card with a soft "paper" surface (mirrors macOS), copy button anchored bottom-right so it no longer overlaps long markdown headings, and an "Answer" tooltip on copy.`,
+            `• Chat: animated typing indicator (pulsing sparkle + three staggered dots) appears the moment you send your first message — matches macOS TypingDotsView.`,
+            `• Chat: extra breathing room between thinking-timeline steps so the agent's intermediate reasoning reads as discrete actions instead of a cramped wall.`,
+            `• Markdown: brand-new Nord-themed renderer — no more white-background leaks from the underlying MdXaml engine on paragraphs, blockquotes, lists, tables, or inline code.`,
+            `• Markdown: bullets and numbered lists are no longer clipped on the left edge.`,
+            `• Markdown: inline code now renders as a soft pill (BadgeFill) instead of a dark slab; fenced code blocks keep their rounded macOS-style chrome with language label + copy.`,
+            `• MCP Servers: editor now supports custom HTTP headers — one Key: Value per line, monospace input, persisted alongside the URL. Authorization headers are unredacted on edit so they round-trip cleanly, and stale fetches won't clobber what you're typing.`,
+            `• Composer: capped + centered at 820 DIP on wide monitors for a balanced layout, full pane width on smaller windows.`,
+            `• Theme: shared interactive-surface brushes (Hover, Press, CodeBackground, UserBubble, AssistantText, DangerSoft, FinalAnswerSurface, BadgeFill) promoted to NordTheme.xaml so every page stays in visual lockstep.`,
         ].join('\n'),
     });
 });

package/backend-dist/mcpServerRoutes.js

@@ -122,7 +122,12 @@ function mcpServerRouter() {
                 return res.status(400).json({ error: 'Invalid MCP server data.' });
             }
             if (err?.name === 'SequelizeUniqueConstraintError') {
-                return res.status(409).json({ error: 'An MCP server with that name already exists.' });
+                const isNameConflict = err?.fields?.includes('name') || err?.errors?.some((e) => e.path === 'name');
+                return res.status(409).json({
+                    error: isNameConflict
+                        ? 'An MCP server with that name already exists.'
+                        : 'Failed to create MCP server due to a conflict.',
+                });
             }
             res.status(500).json({ error: 'Failed to create MCP server.' });
         }

package/package.json

@@ -4,7 +4,7 @@
     "access": "public",
     "registry": "https://registry.npmjs.org/"
   },
-  "version": "1.1.0",
+  "version": "1.3.0",
   "description": "CLI for onboarding users to Omnikey AI and configuring OPENAI_API_KEY. Use Yarn for install/build.",
   "engines": {
     "node": ">=14.0.0",