Add OpenAI Codex OAuth and streaming requests

README.md

@@ -30,8 +30,10 @@
 
 ## Getting Started
 
-1. **Set Keys:**
-   - Add `OPENAI_API_KEY` or `TOGETHER_API_KEY` in `www/config.env`.
+1. **Connect OpenAI (ChatGPT subscription):**
+   - Start the app and click "Connect OpenAI" in the UI to authorize via OAuth.
+   - This uses the same Codex OAuth flow as OpenCode.
+   - Make sure port 1455 is available for the local OAuth callback.
 2. **Problems Setup:**
 
    - Put your Hacker Cup-format problems into `./PROBLEMS/` (see examples)
@@ -54,7 +56,7 @@
 
 5. **Models & Config:**
 
-   - Default model: GPT-4o-mini
+   - Default model: Codex (ChatGPT subscription)
    - See `www/app/config.ts` to:
      - Switch between different LLM models
      - Adjust agent settings and parameters

www/.gitignore

@@ -1,3 +1,4 @@
 .next/ 
 config.env
-node_modules
+node_modules
+.stackfish/

www/app/api/openai/authorize/route.ts

@@ -0,0 +1,12 @@
+import { NextResponse } from 'next/server';
+import { startOpenAIAuthorize } from '../../../services/openaiOAuth';
+
+export async function POST() {
+  try {
+    const authorization = await startOpenAIAuthorize();
+    return NextResponse.json(authorization);
+  } catch (error) {
+    const message = error instanceof Error ? error.message : 'Failed to start OpenAI authorization';
+    return NextResponse.json({ error: message }, { status: 500 });
+  }
+}

www/app/api/openai/logout/route.ts

@@ -0,0 +1,7 @@
+import { NextResponse } from 'next/server';
+import { clearOpenAIAuth } from '../../../services/openaiOAuth';
+
+export async function POST() {
+  await clearOpenAIAuth();
+  return NextResponse.json({ success: true });
+}

www/app/api/openai/status/route.ts

@@ -0,0 +1,9 @@
+import { NextResponse } from 'next/server';
+import { getOpenAIAuthStatus } from '../../../services/openaiOAuth';
+
+export const dynamic = 'force-dynamic';
+
+export async function GET() {
+  const status = await getOpenAIAuthStatus();
+  return NextResponse.json(status);
+}

www/app/config/config.ts

@@ -4,56 +4,66 @@ import { Model } from '../types/models';
 // RECOMMENDED:specify yours CLOUD EXECUTE URL after deploying ./cloud-run-worker to google cloud run
 export const CLOUD_EXECUTE_URL = "https://cloud-run-worker-313568160682.us-central1.run.app/compute";
 
-// If use OpenAI models, please specify OPENAI_API_KEY= in config.env file
+// OpenAI models use ChatGPT subscription OAuth (Codex)
 // If use qwq or llama models, please specify together.ai TOGETHER_API_KEY= in config.env file
 
 // Defines how many parallel LLM calls to launch to generate synthetic tests for a problem
 export const syntheticTestCallsPerModel: Record<Model, number> = {
-  'o1-preview': 0,
-  'o1-mini': 0,
-  'gpt-4o': 0,
-  'gpt-4o-mini': 1,
+  'gpt-5.3-codex': 1,
+  'gpt-5.2-codex': 0,
+  'gpt-5.2': 0,
+  'gpt-5.1-codex': 0,
+  'gpt-5.1-codex-mini': 0,
+  'gpt-5.1-codex-max': 0,
   'qwq-32b-preview': 0,
   'llama-3.3-70b': 0,
 }
 
 export const postSyntheticTestCallsPerModel: Record<Model, number> = {
-  'o1-preview': 0,
-  'o1-mini': 0,
-  'gpt-4o': 0,
-  'gpt-4o-mini': 1,
+  'gpt-5.3-codex': 1,
+  'gpt-5.2-codex': 0,
+  'gpt-5.2': 0,
+  'gpt-5.1-codex': 0,
+  'gpt-5.1-codex-mini': 0,
+  'gpt-5.1-codex-max': 0,
   'qwq-32b-preview': 0,
   'llama-3.3-70b': 0,
 }
 
 // Defines how many parallel LLM calls to launch to generate a hypothesis (aka. attack vector)
 export const attackVectorCallsPerModel: Record<Model, number> = {
-  'o1-preview': 0,
-  'o1-mini': 0,
-  'gpt-4o': 0,
-  'gpt-4o-mini': 1,
+  'gpt-5.3-codex': 1,
+  'gpt-5.2-codex': 0,
+  'gpt-5.2': 0,
+  'gpt-5.1-codex': 0,
+  'gpt-5.1-codex-mini': 0,
+  'gpt-5.1-codex-max': 0,
   'qwq-32b-preview': 0,
   'llama-3.3-70b': 0,
 }
 
 // Defines how many parallel agents to launch to:
 // After hypothesis is generated, how many agents should write a solution code
 export const postAttackVectorSolutionCallsPerModel: Record<Model, number> = {
-  'o1-preview': 0,
-  'o1-mini': 0,
-  'gpt-4o': 0,
-  'gpt-4o-mini': 1,
+  'gpt-5.3-codex': 1,
+  'gpt-5.2-codex': 0,
+  'gpt-5.2': 0,
+  'gpt-5.1-codex': 0,
+  'gpt-5.1-codex-mini': 0,
+  'gpt-5.1-codex-max': 0,
   'qwq-32b-preview': 0,
   'llama-3.3-70b': 0,
 }
 
 // Defines how many parallel agents to launch to:
 // Generate a solution directly, without a hypothesis and tests steps
 export const directSolutionCallsPerModel: Record<Model, number> = {
-  'o1-preview': 0,
-  'o1-mini': 0,
-  'gpt-4o': 0,
-  'gpt-4o-mini': 0,
+  'gpt-5.3-codex': 0,
+  'gpt-5.2-codex': 0,
+  'gpt-5.2': 0,
+  'gpt-5.1-codex': 0,
+  'gpt-5.1-codex-mini': 0,
+  'gpt-5.1-codex-max': 0,
   'qwq-32b-preview': 0,
   'llama-3.3-70b': 0,
 }
@@ -64,14 +74,14 @@ export const directSolutionCallsPerModel: Record<Model, number> = {
 // If only one output is possible, we can verify test cases by simply comparing the strings
 // If more than one output is possible, we need to use LLM to verify if our solution is correct
 // The IS_ONLY_ONE_OUTPUT_VALID_MODEL model reads the problem statement and determines if only one output is possible
-export const IS_ONLY_ONE_OUTPUT_VALID_MODEL: Model = 'gpt-4o';
+export const IS_ONLY_ONE_OUTPUT_VALID_MODEL: Model = 'gpt-5.3-codex';
 
 // In case if multiple outputs are possible, we use LLM to guess if the provided output seems correct
-export const IS_VALID_OUTPUT_MODEL: Model = 'gpt-4o';
+export const IS_VALID_OUTPUT_MODEL: Model = 'gpt-5.3-codex';
 
 // After a hypothesis is generated, we use LLM to extract knowledge tags - advanced algorithms and data structures,
 // which are mentioned in the hypothesis
-export const EXTRACT_KNOWLEDGE_TAGS_MODEL: Model = 'gpt-4o';
+export const EXTRACT_KNOWLEDGE_TAGS_MODEL: Model = 'gpt-5.3-codex';
 
 // A dir with problem statements in hackercup format. Each problem is in a separate dir.
 // Files required in each problem dir: statement.txt, sample_in.txt, sample_out.txt, full_in.txt
@@ -80,4 +90,3 @@ export const PROBLEMS_PATH = path.join(process.cwd(), '..', 'PROBLEMS');
 // A dir where the model will place verified solutions
 // For every problem a separate dir is created, and both source code and full test cases are placed there
 export const SOLUTIONS_PATH = path.join(process.cwd(), '..', 'SOLUTIONS');
-

www/app/page.tsx

@@ -1,13 +1,17 @@
 'use client';
 
-import { useEffect, useState } from 'react';
+import { useEffect, useRef, useState } from 'react';
 import ProblemColumn from '@/components/ProblemColumn';
-import Image from 'next/image';
 import { ProblemService } from './services/problemService';
 
 export default function Home() {
   const [problems, setProblems] = useState<string[]>([]);
   const [problemRequests, setProblemRequests] = useState<Record<string, { llm: number; compute: number }>>({});
+  const [authStatus, setAuthStatus] = useState<{ status: string; connected: boolean; error?: string; accountId?: string }>({
+    status: 'idle',
+    connected: false,
+  });
+  const pollRef = useRef<ReturnType<typeof setInterval> | null>(null);
 
   useEffect(() => {
     const fetchProblems = async () => {
@@ -32,6 +36,56 @@ export default function Home() {
     };
   }, []);
 
+  const refreshAuth = async () => {
+    try {
+      const response = await fetch('/api/openai/status');
+      const data = await response.json();
+      setAuthStatus(data);
+      if (data.status !== 'pending' && pollRef.current) {
+        clearInterval(pollRef.current);
+        pollRef.current = null;
+      }
+    } catch {
+      setAuthStatus((prev) => ({ ...prev, status: 'error', error: 'Failed to read auth status' }));
+    }
+  };
+
+  useEffect(() => {
+    refreshAuth();
+    return () => {
+      if (pollRef.current) {
+        clearInterval(pollRef.current);
+      }
+    };
+  }, []);
+
+  const startAuth = async () => {
+    setAuthStatus((prev) => ({ ...prev, status: 'pending', error: undefined }));
+    try {
+      const response = await fetch('/api/openai/authorize', { method: 'POST' });
+      const data = await response.json();
+      if (data?.error) {
+        setAuthStatus((prev) => ({ ...prev, status: 'error', error: data.error }));
+        return;
+      }
+      if (data?.url) {
+        window.open(data.url, '_blank', 'noopener,noreferrer');
+      }
+      if (!pollRef.current) {
+        pollRef.current = setInterval(() => {
+          refreshAuth();
+        }, 1000);
+      }
+    } catch {
+      setAuthStatus((prev) => ({ ...prev, status: 'error', error: 'Failed to start OAuth' }));
+    }
+  };
+
+  const disconnect = async () => {
+    await fetch('/api/openai/logout', { method: 'POST' });
+    await refreshAuth();
+  };
+
   // Calculate totals
   const totalLLM = Object.values(problemRequests).reduce((sum, curr) => sum + curr.llm, 0);
   const totalCompute = Object.values(problemRequests).reduce((sum, curr) => sum + curr.compute, 0);
@@ -45,7 +99,27 @@ export default function Home() {
               🐟 STACKFISH
             </span>
           </div>
-          <div className="flex gap-2">
+          <div className="flex gap-2 items-center">
+            <div className="flex items-center gap-2 px-4 py-2 rounded-lg border border-white/10">
+              <span className="text-xs font-medium">
+                OpenAI: {authStatus.connected ? 'Connected' : authStatus.status === 'pending' ? 'Connecting' : 'Not connected'}
+              </span>
+              {authStatus.connected ? (
+                <button
+                  className="text-xs px-3 py-1 rounded-md bg-white/10 hover:bg-white/20 transition"
+                  onClick={disconnect}
+                >
+                  Disconnect
+                </button>
+              ) : (
+                <button
+                  className="text-xs px-3 py-1 rounded-md bg-white/10 hover:bg-white/20 transition"
+                  onClick={startAuth}
+                >
+                  Connect OpenAI
+                </button>
+              )}
+            </div>
             <div className="flex items-center gap-2 px-4 py-2 rounded-lg">
               <div className={`w-2 h-2 rounded-full ${totalLLM > 0 ? 'bg-green-500 animate-pulse' : 'bg-gray-500'}`} />
               <span className="text-xs font-medium">