@aj-archipelago/cortex 1.3.23 → 1.3.24

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aj-archipelago/cortex",
-  "version": "1.3.23",
+  "version": "1.3.24",
   "description": "Cortex is a GraphQL API for AI. It provides a simple, extensible interface for using AI services from OpenAI, Azure and others.",
   "private": false,
   "repository": {
@@ -33,7 +33,7 @@
   "type": "module",
   "homepage": "https://github.com/aj-archipelago/cortex#readme",
   "dependencies": {
-    "@aj-archipelago/subvibe": "^1.0.3",
+    "@aj-archipelago/subvibe": "^1.0.5",
     "@apollo/server": "^4.7.3",
     "@apollo/server-plugin-response-cache": "^4.1.2",
     "@apollo/utils.keyvadapter": "^3.0.0",

package/pathways/system/entity/sys_generator_memory.js

@@ -18,7 +18,7 @@ export default {
         const { aiStyle, AI_STYLE_ANTHROPIC, AI_STYLE_OPENAI } = args;
         const styleModel = aiStyle === "Anthropic" ? AI_STYLE_ANTHROPIC : AI_STYLE_OPENAI;
 
-        const memoryContext = await callPathway('sys_search_memory', { ...args, section: 'memoryAll', updateContext: true });
+        const memoryContext = await callPathway('sys_search_memory', { ...args, stream: false, section: 'memoryAll', updateContext: true });
         if (memoryContext) {
             const {toolCallId} = addToolCalls(args.chatHistory, "search memory for relevant information", "memory_lookup");
             addToolResults(args.chatHistory, memoryContext, toolCallId);
@@ -26,9 +26,9 @@ export default {
 
         let result;
         if (args.voiceResponse) {
-            result = await callPathway('sys_generator_quick', { ...args, model: styleModel, stream: false });
+            result = await callPathway('sys_generator_quick', { ...args, model: styleModel, stream: false }, resolver);
         } else {
-            result = await callPathway('sys_generator_quick', { ...args, model: styleModel });
+            result = await callPathway('sys_generator_quick', { ...args, model: styleModel }, resolver);
         }
 
         resolver.tool = JSON.stringify({ toolUsed: "memory" });

package/pathways/system/rest_streaming/sys_openai_chat.js

@@ -13,7 +13,7 @@ export default {
     inputParameters: {
         messages: [],
     },
-    model: 'oai-gpturbo',
+    model: 'oai-gpt4o',
     useInputChunking: false,
-    emulateOpenAIChatModel: 'gpt-3.5-turbo',
+    emulateOpenAIChatModel: 'gpt-4o',
 }

package/server/pathwayResolver.js

@@ -79,6 +79,13 @@ class PathwayResolver {
         let streamErrorOccurred = false;
         let responseData = null;
 
+        const publishNestedRequestProgress = (requestProgress) => {
+            if (requestProgress.progress === 1 && this.rootRequestId) {
+                delete requestProgress.progress;
+            }
+            publishRequestProgress(requestProgress);
+        }
+
         try {
             responseData = await this.executePathway(args);
         }
@@ -105,7 +112,7 @@ class PathwayResolver {
             
             // some models don't support progress updates
             if (!modelTypesExcludedFromProgressUpdates.includes(this.model.type)) {
-                await publishRequestProgress({
+                await publishNestedRequestProgress({
                         requestId: this.rootRequestId || this.requestId,
                         progress: Math.min(completedCount,totalCount) / totalCount,
                         data: JSON.stringify(responseData),
@@ -144,10 +151,7 @@ class PathwayResolver {
 
                     try {
                         if (!streamEnded && requestProgress.data) {
-                            if (!(this.rootRequestId && requestProgress.progress === 1)) {
-                                logger.debug(`Publishing stream message to requestId ${this.requestId}: ${requestProgress.data}`);
-                                publishRequestProgress(requestProgress);
-                            }
+                            publishNestedRequestProgress(requestProgress);
                             streamEnded = requestProgress.progress === 1;
                         }
                     } catch (error) {

package/server/plugins/claude3VertexPlugin.js

@@ -136,7 +136,16 @@ class Claude3VertexPlugin extends OpenAIVisionPlugin {
     // Extract system messages
     const systemMessages = messagesCopy.filter(message => message.role === "system");
     if (systemMessages.length > 0) {
-      system = systemMessages.map(message => message.content).join("\n");
+      system = systemMessages.map(message => {
+        if (Array.isArray(message.content)) {
+          // For content arrays, extract text content and join
+          return message.content
+            .filter(item => item.type === 'text')
+            .map(item => item.text)
+            .join("\n");
+        }
+        return message.content;
+      }).join("\n");
     }
   
     // Filter out system messages and empty messages

package/server/plugins/gemini15ChatPlugin.js

@@ -213,6 +213,10 @@ class Gemini15ChatPlugin extends ModelPlugin {
 
             // If this message also has STOP, mark it for completion but don't overwrite the content
             if (eventData.candidates[0].finishReason === "STOP") {
+                // Send the content first
+                requestProgress.data = JSON.stringify(createChunk({ 
+                    content: eventData.candidates[0].content.parts[0].text
+                }));
                 requestProgress.progress = 1;
             }
         } else if (eventData.candidates?.[0]?.finishReason === "STOP") {

package/tests/openai_api.test.js

@@ -48,7 +48,7 @@ test('POST /completions', async (t) => {
 test('POST /chat/completions', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      model: 'gpt-3.5-turbo',
+      model: 'gpt-4o',
       messages: [{ role: 'user', content: 'Hello!' }],
       stream: false,
     },
@@ -63,7 +63,7 @@ test('POST /chat/completions', async (t) => {
 test('POST /chat/completions with multimodal content', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      model: 'claude-3.5-sonnet',
+      model: 'gpt-4o',
       messages: [{
         role: 'user',
         content: [
@@ -153,7 +153,7 @@ test('POST SSE: /v1/completions should send a series of events and a [DONE] even
 
 test('POST SSE: /v1/chat/completions should send a series of events and a [DONE] event', async (t) => {
     const payload = {
-        model: 'gpt-3.5-turbo',
+        model: 'gpt-4o',
         messages: [
         {
             role: 'user',
@@ -177,7 +177,7 @@ test('POST SSE: /v1/chat/completions should send a series of events and a [DONE]
 
 test('POST SSE: /v1/chat/completions with multimodal content should send a series of events and a [DONE] event', async (t) => {
     const payload = {
-        model: 'claude-3.5-sonnet',
+        model: 'gpt-4o',
         messages: [{
           role: 'user',
           content: [
@@ -213,7 +213,7 @@ test('POST SSE: /v1/chat/completions with multimodal content should send a serie
 test('POST /chat/completions should handle multimodal content for non-multimodal model', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      model: 'gpt-3.5-turbo',
+      model: 'gpt-4o',
       messages: [{
         role: 'user',
         content: [
@@ -242,7 +242,7 @@ test('POST /chat/completions should handle multimodal content for non-multimodal
 
 test('POST SSE: /v1/chat/completions should handle streaming multimodal content for non-multimodal model', async (t) => {
   const payload = {
-    model: 'gpt-3.5-turbo',
+    model: 'gpt-4o',
     messages: [{
       role: 'user',
       content: [
@@ -282,7 +282,7 @@ test('POST SSE: /v1/chat/completions should handle streaming multimodal content
 test('POST /chat/completions should handle malformed multimodal content', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      model: 'claude-3.5-sonnet',
+      model: 'gpt-4o',
       messages: [{
         role: 'user',
         content: [
@@ -310,7 +310,7 @@ test('POST /chat/completions should handle malformed multimodal content', async
 test('POST /chat/completions should handle invalid image data', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      model: 'claude-3.5-sonnet',
+      model: 'gpt-4o',
       messages: [{
         role: 'user',
         content: [
@@ -361,7 +361,7 @@ test('POST /completions should handle model parameters', async (t) => {
 test('POST /chat/completions should handle function calling', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      model: 'gpt-3.5-turbo',
+      model: 'gpt-4o',
       messages: [{ role: 'user', content: 'What is the weather in Boston?' }],
       functions: [{
         name: 'get_weather',
@@ -401,7 +401,7 @@ test('POST /chat/completions should handle function calling', async (t) => {
 test('POST /chat/completions should validate response format', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      model: 'gpt-3.5-turbo',
+      model: 'gpt-4o',
       messages: [{ role: 'user', content: 'Hello!' }],
       stream: false,
     },
@@ -426,7 +426,7 @@ test('POST /chat/completions should validate response format', async (t) => {
 test('POST /chat/completions should handle system messages', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      model: 'gpt-3.5-turbo',
+      model: 'gpt-4o',
       messages: [
         { role: 'system', content: 'You are a helpful assistant.' },
         { role: 'user', content: 'Hello!' }
@@ -443,10 +443,29 @@ test('POST /chat/completions should handle system messages', async (t) => {
 });
 
 test('POST /chat/completions should handle errors gracefully', async (t) => {
+  const error = await t.throwsAsync(
+    () => got.post(`${API_BASE}/chat/completions`, {
+      json: {
+        // Missing required model field
+        messages: [{ role: 'user', content: 'Hello!' }],
+      },
+      responseType: 'json',
+    })
+  );
+  
+  t.is(error.response.statusCode, 404);
+});
+
+test('POST /chat/completions should handle token limits', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      // Missing required model field
-      messages: [{ role: 'user', content: 'Hello!' }],
+      model: 'gpt-4o',
+      messages: [{ 
+        role: 'user', 
+        content: 'Hello!'.repeat(5000) // Very long message
+      }],
+      max_tokens: 100,
+      stream: false,
     },
     responseType: 'json',
   });
@@ -455,17 +474,16 @@ test('POST /chat/completions should handle errors gracefully', async (t) => {
   t.is(response.body.object, 'chat.completion');
   t.true(Array.isArray(response.body.choices));
   t.truthy(response.body.choices[0].message.content);
-});
+});  
 
-test('POST /chat/completions should handle token limits', async (t) => {
+test('POST /chat/completions should return complete responses from gpt-4o', async (t) => {
   const response = await got.post(`${API_BASE}/chat/completions`, {
     json: {
-      model: 'gpt-3.5-turbo',
-      messages: [{ 
-        role: 'user', 
-        content: 'Hello!'.repeat(5000) // Very long message
-      }],
-      max_tokens: 100,
+      model: 'gpt-4o',
+      messages: [
+        { role: 'system', content: 'You are a helpful assistant. Always end your response with the exact string "END_MARKER_XYZ".' },
+        { role: 'user', content: 'Say hello and explain why complete responses matter.' }
+      ],
       stream: false,
     },
     responseType: 'json',
@@ -474,6 +492,8 @@ test('POST /chat/completions should handle token limits', async (t) => {
   t.is(response.statusCode, 200);
   t.is(response.body.object, 'chat.completion');
   t.true(Array.isArray(response.body.choices));
-  t.truthy(response.body.choices[0].message.content);
-});  
+  console.log('GPT-4o Response:', JSON.stringify(response.body.choices[0].message.content));
+  const content = response.body.choices[0].message.content;
+  t.regex(content, /END_MARKER_XYZ$/);
+}); 
   

package/tests/streaming.test.js

@@ -0,0 +1,197 @@
+import test from 'ava';
+import serverFactory from '../index.js';
+import { PathwayResolver } from '../server/pathwayResolver.js';
+import OpenAIChatPlugin from '../server/plugins/openAiChatPlugin.js';
+import GeminiChatPlugin from '../server/plugins/geminiChatPlugin.js';
+import Gemini15ChatPlugin from '../server/plugins/gemini15ChatPlugin.js';
+import Claude3VertexPlugin from '../server/plugins/claude3VertexPlugin.js';
+import { config } from '../config.js';
+
+let testServer;
+
+test.before(async () => {
+  process.env.CORTEX_ENABLE_REST = 'true';
+  const { server, startServer } = await serverFactory();
+  startServer && await startServer();
+  testServer = server;
+});
+
+test.after.always('cleanup', async () => {
+  if (testServer) {
+    await testServer.stop();
+  }
+});
+
+// Helper function to create a PathwayResolver with a specific plugin
+function createResolverWithPlugin(pluginClass, modelName = 'test-model') {
+  // Map plugin classes to their corresponding model types
+  const pluginToModelType = {
+    OpenAIChatPlugin: 'OPENAI-VISION',
+    GeminiChatPlugin: 'GEMINI-VISION',
+    Gemini15ChatPlugin: 'GEMINI-1.5-VISION',
+    Claude3VertexPlugin: 'CLAUDE-3-VERTEX'
+  };
+
+  const modelType = pluginToModelType[pluginClass.name];
+  if (!modelType) {
+    throw new Error(`Unknown plugin class: ${pluginClass.name}`);
+  }
+
+  const pathway = {
+    name: 'test-pathway',
+    model: modelName,
+    prompt: 'test prompt'
+  };
+  
+  const model = {
+    name: modelName,
+    type: modelType
+  };
+  
+  const resolver = new PathwayResolver({ 
+    config,
+    pathway,
+    args: {},
+    endpoints: { [modelName]: model }
+  });
+  
+  resolver.modelExecutor.plugin = new pluginClass(pathway, model);
+  return resolver;
+}
+
+// Test OpenAI Chat Plugin Streaming
+test('OpenAI Chat Plugin - processStreamEvent handles content chunks correctly', async t => {
+  const resolver = createResolverWithPlugin(OpenAIChatPlugin);
+  const plugin = resolver.modelExecutor.plugin;
+  
+  // Test regular content chunk
+  const contentEvent = {
+    data: JSON.stringify({
+      id: 'test-id',
+      choices: [{
+        delta: { content: 'test content' },
+        finish_reason: null
+      }]
+    })
+  };
+  
+  let progress = plugin.processStreamEvent(contentEvent, {});
+  t.is(progress.data, contentEvent.data);
+  t.falsy(progress.progress);
+  
+  // Test stream end
+  const endEvent = {
+    data: JSON.stringify({
+      id: 'test-id',
+      choices: [{
+        delta: {},
+        finish_reason: 'stop'
+      }]
+    })
+  };
+  
+  progress = plugin.processStreamEvent(endEvent, {});
+  t.is(progress.progress, 1);
+});
+
+// Test Gemini Chat Plugin Streaming
+test('Gemini Chat Plugin - processStreamEvent handles content chunks correctly', async t => {
+  const resolver = createResolverWithPlugin(GeminiChatPlugin);
+  const plugin = resolver.modelExecutor.plugin;
+  
+  // Test regular content chunk
+  const contentEvent = {
+    data: JSON.stringify({
+      candidates: [{
+        content: {
+          parts: [{ text: 'test content' }]
+        },
+        finishReason: null
+      }]
+    })
+  };
+  
+  let progress = plugin.processStreamEvent(contentEvent, {});
+  t.truthy(progress.data, 'Should have data');
+  const parsedData = JSON.parse(progress.data);
+  t.truthy(parsedData.candidates, 'Should have candidates array');
+  t.truthy(parsedData.candidates[0].content, 'Should have content object');
+  t.truthy(parsedData.candidates[0].content.parts, 'Should have parts array');
+  t.is(parsedData.candidates[0].content.parts[0].text, 'test content', 'Content should match');
+  t.falsy(progress.progress);
+  
+  // Test stream end with STOP
+  const endEvent = {
+    data: JSON.stringify({
+      candidates: [{
+        content: {
+          parts: [{ text: '' }]
+        },
+        finishReason: 'STOP'
+      }]
+    })
+  };
+  
+  progress = plugin.processStreamEvent(endEvent, {});
+  t.is(progress.progress, 1);
+});
+
+// Test Gemini 15 Chat Plugin Streaming
+test('Gemini 15 Chat Plugin - processStreamEvent handles safety blocks', async t => {
+  const resolver = createResolverWithPlugin(Gemini15ChatPlugin);
+  const plugin = resolver.modelExecutor.plugin;
+  
+  // Test safety block
+  const safetyEvent = {
+    data: JSON.stringify({
+      candidates: [{
+        safetyRatings: [{ blocked: true }]
+      }]
+    })
+  };
+  
+  const progress = plugin.processStreamEvent(safetyEvent, {});
+  t.true(progress.data.includes('Response blocked'));
+  t.is(progress.progress, 1);
+});
+
+// Test Claude 3 Vertex Plugin Streaming
+test('Claude 3 Vertex Plugin - processStreamEvent handles message types', async t => {
+  const resolver = createResolverWithPlugin(Claude3VertexPlugin);
+  const plugin = resolver.modelExecutor.plugin;
+  
+  // Test message start
+  const startEvent = {
+    data: JSON.stringify({
+      type: 'message_start',
+      message: { id: 'test-id' }
+    })
+  };
+  
+  let progress = plugin.processStreamEvent(startEvent, {});
+  t.true(JSON.parse(progress.data).choices[0].delta.role === 'assistant');
+  
+  // Test content block
+  const contentEvent = {
+    data: JSON.stringify({
+      type: 'content_block_delta',
+      delta: {
+        type: 'text_delta',
+        text: 'test content'
+      }
+    })
+  };
+  
+  progress = plugin.processStreamEvent(contentEvent, {});
+  t.true(JSON.parse(progress.data).choices[0].delta.content === 'test content');
+  
+  // Test message stop
+  const stopEvent = {
+    data: JSON.stringify({
+      type: 'message_stop'
+    })
+  };
+  
+  progress = plugin.processStreamEvent(stopEvent, {});
+  t.is(progress.progress, 1);
+});