0.4.0

CONVENTIONS.md

@@ -7,21 +7,3 @@ Use async/await where possible
 Test exceptional cases first and return/throw early.
 
 Never edit files name CONVENTIONS.md or .cursorrules
-
-# Test code standards
-
-Unit test files should be in the same directory as the source file.
-
-Any usage of chai-as-promised should use async/await
-```
-it('should work well with async/await', async () => {
-    (await Promise.resolve(42)).should.equal(42)
-    await Promise.reject(new Error()).should.be.rejectedWith(Error);
-});
-```
-
-# Tool/function classes
-
-Function classes with the @funcClass(__filename) must only have the default constructor.
-
-Always use the Filesystem class in src/functions/storage/filesystem.ts to read/search/write to the local filesystem.

frontend/src/app/modules/agents/agent/agent-details/agent-details.component.html

@@ -1,7 +1,7 @@
 @if(agentDetails) {
     <mat-card *ngIf="agentDetails.state === 'feedback'" class="p-3 mb-4">
         <form [formGroup]="feedbackForm" (ngSubmit)="onSubmitFeedback()">
-            <mat-card-title>Feedback Requested</mat-card-title>
+            <mat-card-title class="font-bold pl-5 text-lg">Feedback Requested</mat-card-title>
             <mat-card-content style="margin-bottom: 0; margin-top: 15px">
                 <mat-expansion-panel
                         *ngIf="agentDetails.functionCallHistory && agentDetails.functionCallHistory.length > 0"
@@ -42,7 +42,7 @@
 
     <mat-card *ngIf="agentDetails?.state === 'error'" class="p-3 mb-4">
         <form [formGroup]="errorForm" (ngSubmit)="onResumeError()">
-            <mat-card-title>Agent Error</mat-card-title>
+            <mat-card-title class="font-bold pl-5 text-lg">Agent Error</mat-card-title>
             <mat-card-content style="margin-bottom: 0; margin-top: 15px">
                 <mat-expansion-panel *ngIf="agentDetails.error" style="width: 1000px; margin-bottom: 20px">
                     <mat-expansion-panel-header>
@@ -75,7 +75,7 @@
 
     <mat-card *ngIf="agentDetails?.state === 'hil'" class="p-3 mb-4">
         <form [formGroup]="hilForm" (ngSubmit)="onResumeHil()">
-            <mat-card-title>Human In Loop check</mat-card-title>
+            <mat-card-title class="font-bold pl-5 text-lg">Human In Loop check</mat-card-title>
             <mat-card-content>
                 <mat-form-field appearance="fill" class="full-width">
                     <textarea

frontend/src/app/modules/agents/agent/agent-details/agent-details.component.ts

@@ -123,6 +123,7 @@ export class AgentDetailsComponent implements OnInit {
             })
         ).subscribe((response) => {
             if (response) {
+                this.feedbackForm.reset();
                 this.snackBar.open('Feedback submitted successfully', 'Close', { duration: 3000 });
                 this.refreshAgentDetails();
             }
@@ -289,9 +290,7 @@ export class AgentDetailsComponent implements OnInit {
         });
 
         dialogRef.afterClosed().subscribe((result) => {
-            if (result) {
-                this.resumeCompletedAgent(result.resumeInstructions);
-            }
+            if (result) this.resumeCompletedAgent(result.resumeInstructions);
         });
     }
 

frontend/src/app/modules/agents/agent/agent-function-calls/agent-function-calls.component.ts

@@ -12,7 +12,20 @@ import { MatExpansionModule } from '@angular/material/expansion';
                 <div class="mb-3 font-medium text-xl">{{ invoked.function_name }}</div>
 
                 <div *ngFor="let param of invoked.parameters | keyvalue">
-                    <p><strong>{{ param.key }}:</strong> {{ param.value }}</p>
+                    <div>
+                        <strong>{{ param.key }}:</strong>
+                        <ng-container *ngIf="param.value?.toString().length <= 200">
+                            {{ param.value }}
+                        </ng-container>
+                        <mat-expansion-panel *ngIf="param.value?.toString().length > 200" class="mt-4" #expansionPanel>
+                            <mat-expansion-panel-header [class.expanded-header]="expansionPanel.expanded">
+                                <mat-panel-title class="font-normal" *ngIf="!expansionPanel.expanded">
+                                    {{ param.value?.toString().substring(0, 200) }}...
+                                </mat-panel-title>
+                            </mat-expansion-panel-header>
+                            <p>{{ param.value }}</p>
+                        </mat-expansion-panel>
+                    </div>
                 </div>
                 <mat-expansion-panel *ngIf="invoked.stdout" class="mt-4">
                     <mat-expansion-panel-header>
@@ -29,6 +42,7 @@ import { MatExpansionModule } from '@angular/material/expansion';
             </div>
         </mat-card>
     `,
+    styles: `.mat-expansion-panel-header.mat-expanded.expanded-header {  height: 1.3em; padding-top: 1.2em; }`,
     standalone: true,
     imports: [CommonModule, MatCardModule, MatExpansionModule],
 })

frontend/src/app/modules/agents/new-agent/new-agent.component.ts

@@ -177,7 +177,7 @@ export class NewAgentComponent implements OnInit {
         .subscribe({
           next: (response) => {
             this.snackBar.open('Agent started', 'Close', { duration: 3000 });
-            this.router.navigate(['/ui/agent', response.data.agentId]).catch(console.error);
+            this.router.navigate(['/ui/agents', response.data.agentId]).catch(console.error);
           },
           error: (error) => {
             this.snackBar.open(`Error ${error.message}`, 'Close', { duration: 3000 });

src/CONVENTIONS.md

@@ -0,0 +1,19 @@
+# Backend code standards
+
+## Test code standards
+
+Unit test files should be in the same directory as the source file.
+
+Any usage of chai-as-promised should use async/await
+```
+it('should work well with async/await', async () => {
+    (await Promise.resolve(42)).should.equal(42)
+    await Promise.reject(new Error()).should.be.rejectedWith(Error);
+});
+```
+
+## Tool/function classes
+
+Function classes with the @funcClass(__filename) must only have the default constructor.
+
+Always use the Filesystem class in src/functions/storage/filesystem.ts to read/search/write to the local filesystem.

src/agent/cachingCodeGenAgentRunner.ts

@@ -259,7 +259,7 @@ export async function runCachingCodegenAgent(agent: AgentContext): Promise<Agent
 						.filter((pkg) => llmPythonCode.includes(`${pkg}.`) || pkg === 'json') // always need json for JsProxyEncoder
 						.map((pkg) => `import ${pkg}\n`)
 						.join();
-					logger.info(`Allowed imports: ${pythonScript}`);
+
 					pythonScript += `
 from typing import Any, List, Dict, Tuple, Optional, Union
 from pyodide.ffi import JsProxy

src/agent/codeGenAgentRunner.ts

@@ -3,7 +3,13 @@ import { Span, SpanStatusCode } from '@opentelemetry/api';
 import { PyodideInterface, loadPyodide } from 'pyodide';
 import { runAgentCompleteHandler } from '#agent/agentCompletion';
 import { AgentContext } from '#agent/agentContextTypes';
-import { AGENT_COMPLETED_NAME, AGENT_REQUEST_FEEDBACK, AGENT_SAVE_MEMORY_CONTENT_PARAM_NAME } from '#agent/agentFunctions';
+import {
+	AGENT_COMPLETED_NAME,
+	AGENT_COMPLETED_PARAM_NAME,
+	AGENT_REQUEST_FEEDBACK,
+	AGENT_SAVE_MEMORY_CONTENT_PARAM_NAME,
+	REQUEST_FEEDBACK_PARAM_NAME,
+} from '#agent/agentFunctions';
 import { buildFunctionCallHistoryPrompt, buildMemoryPrompt, buildToolStatePrompt, updateFunctionSchemas } from '#agent/agentPromptUtils';
 import { AgentExecution, formatFunctionError, formatFunctionResult } from '#agent/agentRunner';
 import { convertJsonToPythonDeclaration, extractPythonCode } from '#agent/codeGenAgentUtils';
@@ -215,7 +221,7 @@ export async function runCodeGenAgent(agent: AgentContext): Promise<AgentExecuti
 						.filter((pkg) => llmPythonCode.includes(`${pkg}.`) || pkg === 'json') // always need json for JsProxyEncoder
 						.map((pkg) => `import ${pkg}\n`)
 						.join();
-					logger.info(`Allowed imports: ${pythonScript}`);
+
 					pythonScript += `
 from typing import Any, List, Dict, Tuple, Optional, Union
 from pyodide.ffi import JsProxy
@@ -265,18 +271,18 @@ main()`.trim();
 
 						// Should force completed/requestFeedback to exit the script - throw a particular Error class
 						if (lastFunctionCall.function_name === AGENT_COMPLETED_NAME) {
-							logger.info('Task completed');
+							logger.info(`Task completed: ${lastFunctionCall.parameters[AGENT_COMPLETED_PARAM_NAME]}`);
 							agent.state = 'completed';
 							completed = true;
 						} else if (lastFunctionCall.function_name === AGENT_REQUEST_FEEDBACK) {
-							logger.info('Feedback requested');
+							logger.info(`Feedback requested: ${lastFunctionCall.parameters[REQUEST_FEEDBACK_PARAM_NAME]}`);
 							agent.state = 'feedback';
 							requestFeedback = true;
 						} else {
 							if (!anyFunctionCallErrors && !completed && !requestFeedback) agent.state = 'agent';
 						}
 					} catch (e) {
-						logger.info(`Caught function error ${e.message}`);
+						logger.info(e, `Caught function error ${e.message}`);
 						functionErrorCount++;
 					}
 					// Function invocations are complete

src/agent/forceStopAgent.ts

@@ -0,0 +1,52 @@
+import { agentContext, agentContextStorage } from '#agent/agentContextLocalStorage';
+import { isExecuting } from '#agent/agentContextTypes';
+import { currentUser } from '#user/userService/userContext';
+import { appContext } from '../applicationContext';
+
+const agentsToStop = new Set<string>();
+
+/**
+ * Terminates the execution of an agent as soon as possible.
+ * @param agentId
+ */
+export async function forceStopAgent(agentId: string): Promise<void> {
+	const agent = await appContext().agentStateService.load(agentId);
+	if (!agent) throw new Error(`No agent with id ${agentId}`);
+	if (!isExecuting(agent)) throw new Error(`Agent ${agentId} is not in an executing state`);
+	if (agent.user.id !== currentUser().id) throw new Error('Cannot stop an agent owned by another user');
+
+	agentsToStop.add(agent.agentId);
+
+	// Reload the agent every 5 seconds for up to a minute and see if it's not in an executing state
+	return new Promise((resolve, reject) => {
+		const startTime = Date.now();
+		const interval = setInterval(async () => {
+			const updatedAgent = await appContext().agentStateService.load(agentId);
+			// Agent should be in an error state if the checkForceStopped() function has been called in its execution
+			if (!isExecuting(updatedAgent)) {
+				clearInterval(interval);
+				agentsToStop.delete(agent.agentId);
+				resolve();
+			} else if (Date.now() - startTime >= 60000) {
+				// 1 minute timeout
+				clearInterval(interval);
+				agentsToStop.delete(agent.agentId);
+				reject(new Error(`Agent ${agentId} did not stop executing within 1 minute`));
+			}
+		}, 5000);
+	});
+}
+
+/**
+ * Checks if the current agent should be force stopped
+ */
+export function checkForceStopped(): void {
+	const agent = agentContext();
+	if (!agent) return;
+	const agentId = typeof agent === 'string' ? agent : agent.agentId;
+
+	if (agentsToStop.has(agentId)) {
+		agentsToStop.delete(agentId);
+		throw new Error(`Agent ${agentId} has been force stopped by user ${currentUser().id}`);
+	}
+}

src/cache/cacheRetry.ts

@@ -40,7 +40,7 @@ export function cacheRetry(options: Partial<CacheRetryOptions> = DEFAULTS) {
 			const cacheRetryOptions = optionsWithDefaults(options);
 
 			const cacheService = appContext().functionCacheService;
-			// console.log(this.constructor.name, methodName, args)
+
 			if (cacheRetryOptions.scope) {
 				const cachedValue = await cacheService.getValue(cacheRetryOptions.scope, this.constructor.name, methodName, args);
 
@@ -51,7 +51,7 @@ export function cacheRetry(options: Partial<CacheRetryOptions> = DEFAULTS) {
 			}
 
 			for (let attempt = 1; attempt <= cacheRetryOptions.retries; attempt++) {
-				logger.debug(`${this.constructor.name}${FUNC_SEP}${methodName} retry ${attempt - 1}`);
+				if (attempt > 1) logger.debug(`${this.constructor.name}${FUNC_SEP}${methodName} retry ${attempt - 1}`);
 				try {
 					let result = originalMethod.apply(this, args);
 					if (typeof result?.then === 'function') result = await result;

src/cli/gaia.ts

@@ -11,11 +11,12 @@ import { PublicWeb } from '#functions/web/web';
 import { LlmCall } from '#llm/llmCallService/llmCall';
 import { ClaudeLLMs } from '#llm/services/anthropic';
 import { Claude3_5_Sonnet_Vertex, ClaudeVertexLLMs } from '#llm/services/anthropic-vertex';
-import { groqLlama3_1_70B } from '#llm/services/groq';
+import { groqLlama3_3_70B } from '#llm/services/groq';
 import { Gemini_1_5_Flash } from '#llm/services/vertexai';
 import { logger } from '#o11y/logger';
 import { sleep } from '#utils/async-utils';
 
+import { openAIo1 } from '#llm/services/openai';
 import { appContext, initFirestoreApplicationContext } from '../applicationContext';
 
 const SYSTEM_PROMPT = `Finish your answer with the following template: FINAL ANSWER: [YOUR FINAL ANSWER]. YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.`;
@@ -92,9 +93,9 @@ async function answerGaiaQuestion(task: GaiaQuestion): Promise<GaiaResult> {
 			// llms: ClaudeVertexLLMs(),
 			llms: {
 				easy: Gemini_1_5_Flash(),
-				medium: groqLlama3_1_70B(),
+				medium: groqLlama3_3_70B(),
 				hard: Claude3_5_Sonnet_Vertex(),
-				xhard: Claude3_5_Sonnet_Vertex(),
+				xhard: openAIo1(),
 			},
 			agentName: `gaia-${task.task_id}`,
 			type: 'codegen',

src/cli/gen.ts

@@ -7,9 +7,9 @@ import { LlmFunctions } from '#agent/LlmFunctions';
 import { agentContext, agentContextStorage, createContext } from '#agent/agentContextLocalStorage';
 import { AgentContext, AgentLLMs } from '#agent/agentContextTypes';
 import { LLM } from '#llm/llm';
-import { ClaudeLLMs } from '#llm/models/anthropic';
-import { Claude3_5_Sonnet_Vertex, ClaudeVertexLLMs } from '#llm/models/anthropic-vertex';
-import { GPT4oMini } from '#llm/models/openai';
+import { ClaudeLLMs } from '#llm/services/anthropic';
+import { Claude3_5_Sonnet_Vertex, ClaudeVertexLLMs } from '#llm/services/anthropic-vertex';
+import { GPT4oMini, openAIo1 } from '#llm/services/openai';
 import { currentUser } from '#user/userService/userContext';
 import { initFirestoreApplicationContext } from '../applicationContext';
 import { CliOptions, getLastRunAgentId, parseProcessArgs, saveAgentId } from './cli';

src/cli/query.ts

@@ -9,7 +9,6 @@ import { ClaudeLLMs } from '#llm/services/anthropic';
 import { ClaudeVertexLLMs } from '#llm/services/anthropic-vertex';
 import { cerebrasLlama3_70b } from '#llm/services/cerebras';
 import { deepseekChat } from '#llm/services/deepseek';
-import { groqLlama3_1_70B } from '#llm/services/groq';
 import { GPT4oMini, openAIo1, openAIo1mini } from '#llm/services/openai';
 import { Gemini_1_5_Flash } from '#llm/services/vertexai';
 import { codebaseQuery } from '#swe/discovery/codebaseQuery';

src/cli/swebench.ts

@@ -14,7 +14,6 @@ import { PublicWeb } from '#functions/web/web';
 import { LlmCall } from '#llm/llmCallService/llmCall';
 import { ClaudeLLMs } from '#llm/services/anthropic';
 import { Claude3_5_Sonnet_Vertex, ClaudeVertexLLMs } from '#llm/services/anthropic-vertex';
-import { groqLlama3_1_70B } from '#llm/services/groq';
 import { Gemini_1_5_Flash } from '#llm/services/vertexai';
 import { logger } from '#o11y/logger';
 import { SWEBenchAgent, SWEInstance } from '#swe/SWEBenchAgent';

src/cli/watch.ts

@@ -24,14 +24,14 @@ export function startWatcher() {
 	const watcher = fs.watch(watchPath, { recursive: true }, async (event: WatchEventType, filename: string | null) => {
 		// Early exit if filename is null
 		if (!filename) return;
-		console.log(filename);
-		console.log(event);
+		console.log(`${event} ${filename}`);
+
 		const filePath = path.join(process.cwd(), watchPath, filename);
 		if (!fileExistsSync(filePath)) {
 			logger.debug(`${filePath} doesn't exist`);
 			return;
 		}
-		logger.info(`Checking ${filePath}`);
+		console.log(`Checking ${filePath}`);
 		try {
 			const data = await fs.promises.readFile(filePath, 'utf-8');
 
@@ -44,7 +44,7 @@ export function startWatcher() {
 			const lines = data.split('\n');
 
 			// Find the index of the first line that starts with '//>>' and ends with '//'
-			const index = lines.findIndex((line) => line.includes('//>>') && line.trim().endsWith('//'));
+			const index = lines.findIndex((line) => line.includes('//>') && line.trim().endsWith('//'));
 
 			// Early exit if no matching lines are found
 			if (index === -1) return;
@@ -54,7 +54,7 @@ export function startWatcher() {
 			const indentation = line.match(/^\s*/)[0]; // Capture leading whitespace for indentation
 			const requirements = line.trim().slice(3, -2).trim();
 
-			logger.info(requirements);
+			logger.info(`Extracted requirements: ${requirements}`);
 
 			// Formulate the prompt
 			const prompt = `You are to implement the TODO instructions on the line which starts with //>> and ends with //.\ni.e: ${requirements}`;

src/fastify/trace-init/trace-init.ts

@@ -10,6 +10,7 @@ import { PinoInstrumentation } from './instrumentation';
 
 import { SemanticResourceAttributes } from '@opentelemetry/semantic-conventions';
 import { agentContextStorage } from '#agent/agentContextLocalStorage';
+import { checkForceStopped } from '#agent/forceStopAgent';
 import { setTracer } from '#o11y/trace';
 
 let initialized = false;
@@ -104,9 +105,9 @@ function initTrace(): void {
 		});
 
 		const tracer = trace.getTracer(traceServiceName);
-		setTracer(tracer, agentContextStorage);
+		setTracer(tracer, agentContextStorage, checkForceStopped);
 	} else {
-		setTracer(null, agentContextStorage);
+		setTracer(null, agentContextStorage, checkForceStopped);
 	}
 }
 

src/functionRegistry.ts

@@ -1,4 +1,4 @@
-import { GoogleCloud } from '#functions/cloud/google-cloud';
+import { GoogleCloud } from '#functions/cloud/google/google-cloud';
 import { ImageGen } from '#functions/image';
 import { Jira } from '#functions/jira';
 import { GitHub } from '#functions/scm/github';