onbuzz 4.8.2 → 4.8.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "onbuzz",
-  "version": "4.8.2",
+  "version": "4.8.3",
   "description": "Loxia OnBuzz - Your AI Fleet",
   "type": "module",
   "main": "src/index.js",

package/src/services/__tests__/compactionRetry.test.js

@@ -19,11 +19,11 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
   };
 
   const mockModelsService = {
-    getAvailableModelNames: jest.fn().mockReturnValue(['gpt-5.1-codex-mini', 'gpt-5-mini']),
+    getAvailableModelNames: jest.fn().mockReturnValue(['gpt-4.1-nano', 'gpt-4.1-mini']),
     getModels: jest.fn().mockReturnValue([
-      { name: 'gpt-5.1-codex-mini', type: 'chat', contextWindow: 400000 },
-      { name: 'gpt-5-mini', type: 'chat', contextWindow: 400000 },
-      { name: 'gpt-5-nano', type: 'chat', contextWindow: 400000 },
+      { name: 'gpt-4.1-nano', type: 'chat', contextWindow: 400000 },
+      { name: 'gpt-4.1-mini', type: 'chat', contextWindow: 400000 },
+      { name: 'gpt-4.1-nano', type: 'chat', contextWindow: 400000 },
       { name: 'random-model-xyz', type: 'chat', contextWindow: 200000 }
     ])
   };
@@ -53,12 +53,12 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
       const onRetryAttempt = jest.fn();
 
       // Only 2 validated models available; first fails, second succeeds
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini', 'gpt-5-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano', 'gpt-4.1-mini']);
       mockAiService.sendMessage
         .mockRejectedValueOnce(new Error('Service unavailable'))
         .mockResolvedValueOnce({ content: 'Summary of conversation' });
 
-      await service._generateSummary(testMessages, 'gpt-5.1-codex-mini', { onRetryAttempt });
+      await service._generateSummary(testMessages, 'gpt-4.1-nano', { onRetryAttempt });
 
       expect(onRetryAttempt).toHaveBeenCalledTimes(1);
     });
@@ -66,18 +66,18 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
     test('onRetryAttempt receives correct message, failedModel, nextModel, attempt', async () => {
       const onRetryAttempt = jest.fn();
 
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini', 'gpt-5-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano', 'gpt-4.1-mini']);
       mockAiService.sendMessage
         .mockRejectedValueOnce(new Error('Service unavailable'))
         .mockResolvedValueOnce({ content: 'Summary of conversation' });
 
-      await service._generateSummary(testMessages, 'gpt-5.1-codex-mini', { onRetryAttempt });
+      await service._generateSummary(testMessages, 'gpt-4.1-nano', { onRetryAttempt });
 
       expect(onRetryAttempt).toHaveBeenCalledWith(
         expect.objectContaining({
           type: 'compaction_retry',
-          failedModel: 'gpt-5.1-codex-mini',
-          nextModel: 'gpt-5-mini',
+          failedModel: 'gpt-4.1-nano',
+          nextModel: 'gpt-4.1-mini',
           attempt: 1
         })
       );
@@ -87,14 +87,14 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
       const onRetryAttempt = jest.fn();
 
       // Only 1 validated model, and no suitable random models
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano']);
       mockModelsService.getModels.mockReturnValue([
-        { name: 'gpt-5.1-codex-mini', type: 'chat', contextWindow: 400000 }
+        { name: 'gpt-4.1-nano', type: 'chat', contextWindow: 400000 }
       ]);
       mockAiService.sendMessage.mockRejectedValue(new Error('Service unavailable'));
 
       await expect(
-        service._generateSummary(testMessages, 'gpt-5.1-codex-mini', { onRetryAttempt })
+        service._generateSummary(testMessages, 'gpt-4.1-nano', { onRetryAttempt })
       ).rejects.toThrow();
 
       // The only call to onRetryAttempt would be from the last-resort block, but
@@ -108,16 +108,16 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
 
   describe('Random model fallback', () => {
     test('after all recommended models fail, tries a random model from modelsService', async () => {
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano']);
       mockModelsService.getModels.mockReturnValue([
-        { name: 'gpt-5.1-codex-mini', type: 'chat', contextWindow: 400000 },
+        { name: 'gpt-4.1-nano', type: 'chat', contextWindow: 400000 },
         { name: 'random-model-xyz', type: 'chat', contextWindow: 200000 }
       ]);
       mockAiService.sendMessage
         .mockRejectedValueOnce(new Error('Service unavailable'))  // recommended model fails
         .mockResolvedValueOnce({ content: 'Last-resort summary' }); // random model succeeds
 
-      const result = await service._generateSummary(testMessages, 'gpt-5.1-codex-mini', {});
+      const result = await service._generateSummary(testMessages, 'gpt-4.1-nano', {});
 
       // Should have been called twice: once for recommended, once for random
       expect(mockAiService.sendMessage).toHaveBeenCalledTimes(2);
@@ -125,16 +125,16 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
     });
 
     test('random model success returns valid summary and does not throw', async () => {
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano']);
       mockModelsService.getModels.mockReturnValue([
-        { name: 'gpt-5.1-codex-mini', type: 'chat', contextWindow: 400000 },
+        { name: 'gpt-4.1-nano', type: 'chat', contextWindow: 400000 },
         { name: 'fallback-model', type: 'chat', contextWindow: 200000 }
       ]);
       mockAiService.sendMessage
         .mockRejectedValueOnce(new Error('Service unavailable'))
         .mockResolvedValueOnce({ content: 'Fallback summary content' });
 
-      const result = await service._generateSummary(testMessages, 'gpt-5.1-codex-mini', {});
+      const result = await service._generateSummary(testMessages, 'gpt-4.1-nano', {});
 
       expect(result.role).toBe('system');
       expect(result.type).toBe('summary');
@@ -142,44 +142,44 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
     });
 
     test('random model failure still throws ALL_MODELS_EXHAUSTED', async () => {
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano']);
       mockAiService.sendMessage.mockRejectedValue(new Error('Everything is broken'));
 
       await expect(
-        service._generateSummary(testMessages, 'gpt-5.1-codex-mini', {})
+        service._generateSummary(testMessages, 'gpt-4.1-nano', {})
       ).rejects.toThrow('ALL_MODELS_EXHAUSTED');
     });
 
     test('random model is NOT one already attempted (filtered out)', async () => {
-      // Only gpt-5.1-codex-mini is validated; random pool has others
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini']);
+      // Only gpt-4.1-nano is validated; random pool has others
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano']);
       mockModelsService.getModels.mockReturnValue([
-        { name: 'gpt-5.1-codex-mini', type: 'chat', contextWindow: 400000 },
+        { name: 'gpt-4.1-nano', type: 'chat', contextWindow: 400000 },
         { name: 'random-model-xyz', type: 'chat', contextWindow: 200000 }
       ]);
       mockAiService.sendMessage
-        .mockRejectedValueOnce(new Error('fail'))  // gpt-5.1-codex-mini fails
+        .mockRejectedValueOnce(new Error('fail'))  // gpt-4.1-nano fails
         .mockResolvedValueOnce({ content: 'Random success' }); // random-model-xyz succeeds
 
-      const result = await service._generateSummary(testMessages, 'gpt-5.1-codex-mini', {});
+      const result = await service._generateSummary(testMessages, 'gpt-4.1-nano', {});
 
       // Second call should be the random model, not the already-attempted one
       const secondCallModel = mockAiService.sendMessage.mock.calls[1][0];
-      expect(secondCallModel).not.toBe('gpt-5.1-codex-mini');
-      expect(result.metadata.compactionModel).not.toBe('gpt-5.1-codex-mini');
+      expect(secondCallModel).not.toBe('gpt-4.1-nano');
+      expect(result.metadata.compactionModel).not.toBe('gpt-4.1-nano');
     });
 
     test('random model must have sufficient context window', async () => {
       // All models except the recommended one have tiny context windows
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano']);
       mockModelsService.getModels.mockReturnValue([
-        { name: 'gpt-5.1-codex-mini', type: 'chat', contextWindow: 400000 },
+        { name: 'gpt-4.1-nano', type: 'chat', contextWindow: 400000 },
         { name: 'tiny-model', type: 'chat', contextWindow: 100 } // too small
       ]);
       mockAiService.sendMessage.mockRejectedValue(new Error('fail'));
 
       await expect(
-        service._generateSummary(testMessages, 'gpt-5.1-codex-mini', {})
+        service._generateSummary(testMessages, 'gpt-4.1-nano', {})
       ).rejects.toThrow('ALL_MODELS_EXHAUSTED');
 
       // Should only have tried the recommended model, not the tiny one
@@ -193,11 +193,11 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
     test('onAllModelsExhausted is called only after ALL models (including random) fail', async () => {
       const onAllModelsExhausted = jest.fn();
 
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano']);
       mockAiService.sendMessage.mockRejectedValue(new Error('fail'));
 
       await expect(
-        service._generateSummary(testMessages, 'gpt-5.1-codex-mini', { onAllModelsExhausted })
+        service._generateSummary(testMessages, 'gpt-4.1-nano', { onAllModelsExhausted })
       ).rejects.toThrow('ALL_MODELS_EXHAUSTED');
 
       expect(onAllModelsExhausted).toHaveBeenCalledTimes(1);
@@ -211,16 +211,16 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
     test('onAllModelsExhausted includes all attempted model names', async () => {
       const onAllModelsExhausted = jest.fn();
 
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini', 'gpt-5-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano', 'gpt-4.1-mini']);
       mockAiService.sendMessage.mockRejectedValue(new Error('fail'));
 
       await expect(
-        service._generateSummary(testMessages, 'gpt-5.1-codex-mini', { onAllModelsExhausted })
+        service._generateSummary(testMessages, 'gpt-4.1-nano', { onAllModelsExhausted })
       ).rejects.toThrow('ALL_MODELS_EXHAUSTED');
 
       const callArg = onAllModelsExhausted.mock.calls[0][0];
-      expect(callArg.models).toContain('gpt-5.1-codex-mini');
-      expect(callArg.models).toContain('gpt-5-mini');
+      expect(callArg.models).toContain('gpt-4.1-nano');
+      expect(callArg.models).toContain('gpt-4.1-mini');
       // Should also include at least one random model that was attempted
       expect(callArg.models.length).toBeGreaterThanOrEqual(2);
     });
@@ -233,10 +233,10 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
       const onRetryAttempt = jest.fn();
       const onAllModelsExhausted = jest.fn();
 
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini', 'gpt-5-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano', 'gpt-4.1-mini']);
       mockAiService.sendMessage.mockResolvedValueOnce({ content: 'Great summary here' });
 
-      const result = await service._generateSummary(testMessages, 'gpt-5.1-codex-mini', {
+      const result = await service._generateSummary(testMessages, 'gpt-4.1-nano', {
         onRetryAttempt,
         onAllModelsExhausted
       });
@@ -253,12 +253,12 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
       const onRetryAttempt = jest.fn();
       const onAllModelsExhausted = jest.fn();
 
-      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-5.1-codex-mini', 'gpt-5-mini']);
+      mockModelsService.getAvailableModelNames.mockReturnValue(['gpt-4.1-nano', 'gpt-4.1-mini']);
       mockAiService.sendMessage
         .mockRejectedValueOnce(new Error('429 rate limit'))
         .mockResolvedValueOnce({ content: 'Second model summary' });
 
-      const result = await service._generateSummary(testMessages, 'gpt-5.1-codex-mini', {
+      const result = await service._generateSummary(testMessages, 'gpt-4.1-nano', {
         onRetryAttempt,
         onAllModelsExhausted
       });
@@ -266,7 +266,7 @@ describe('ConversationCompactionService - _generateSummary retry behavior', () =
       expect(onRetryAttempt).toHaveBeenCalledTimes(1);
       expect(onAllModelsExhausted).not.toHaveBeenCalled();
       expect(result.content).toContain('Second model summary');
-      expect(result.metadata.compactionModel).toBe('gpt-5-mini');
+      expect(result.metadata.compactionModel).toBe('gpt-4.1-mini');
     });
   });
 });

package/src/utilities/constants.js

@@ -656,22 +656,30 @@ const COMPACTION_CONFIG = {
   MIN_MIDDLE_SEGMENT_PERCENTAGE: 0.50, // Middle must be at least 50% of messages
   MAX_BOOKEND_PERCENTAGE: 0.50, // Beginning + end together capped at 50%
 
-  // Recommended model pool for compaction (validated against live model catalog at runtime)
-  // Names MUST match catalog keys exactly (no azure-openai- prefix)
-  // Ordered by context window size (largest first) to handle very large conversations
+  // Recommended model pool for compaction (validated against live model catalog at runtime).
+  // Names MUST match catalog keys exactly (no azure-openai- prefix).
+  //
+  // Compaction is a structured transcription job, not a reasoning task — the
+  // P7 prompt's three passes (USER VOICE / EVENT LOG / STATE NARRATIVE)
+  // produce ~1.5K output tokens of well-shaped text. Using reasoning models
+  // for this burns reasoning-token cost on a task that doesn't need it AND
+  // adds significant TTFT/total latency per pass. The previous defaults
+  // (gpt-5.1-codex-mini, gpt-5-mini, gpt-5-nano, o4-mini — all reasoning)
+  // were chosen for their 400K context windows, but the gpt-4.1 family
+  // offers 1M-token context AND is non-reasoning AND ~30× cheaper than the
+  // previous primary.
+  //
+  // Both entries below have 1M context, so the runtime model-filter never
+  // has to fall through to a smaller-context model for size reasons.
   COMPACTION_MODELS: [
-    'gpt-5.1-codex-mini',           // 400K context - best for large conversations
-    'gpt-5-mini',                   // 400K context
-    'gpt-5-nano',                   // 400K context - lightweight
-    'o4-mini'                       // 128K context - reasoning model
+    'gpt-4.1-nano',                 // 1M context, non-reasoning, cheapest — primary
+    'gpt-4.1-mini',                 // 1M context, non-reasoning           — fallback
   ],
 
   // Context windows for recommended compaction models (fallback if modelsService unavailable)
   MODEL_CONTEXT_WINDOWS: {
-    'gpt-5.1-codex-mini': 400000,
-    'gpt-5-mini': 400000,
-    'gpt-5-nano': 400000,
-    'o4-mini': 128000
+    'gpt-4.1-nano': 1048576,
+    'gpt-4.1-mini': 1048576,
   },
 
   // Token limits