fix: replace AiGateway.embedBatch with direct LiteLLM cURL for upload indexing

AiGateway uses getenv(LITELLM_MASTER_KEY) + stream_context HTTP which was
failing on the chloe virtualhost process. New dbnToolsLiteLLMEmbedBatch()
helper mirrors dbnToolsCallGpuLlm — hardcoded URL + key, cURL-first, same
pattern already proven for LLM calls. Removes AiGateway dependency from
DeepResearchAgent entirely.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

includes/DeepResearchAgent.php

@@ -14,7 +14,6 @@ final class DbnDeepResearchAgent
     private const POOL_CAP = 30;
 
     private DbnAzureOpenAiGateway $azure;
-    private ?AiGateway $ai = null;
     private array $uploadVecs = [];
     private array $stepTimings = [];
 
@@ -50,9 +49,7 @@ final class DbnDeepResearchAgent
         dbnToolsBootCaveau();
         $aiPortalRoot = dbnToolsAiPortalRoot();
         require_once $aiPortalRoot . '/platform/includes/dbn_v6.php';
-        require_once $aiPortalRoot . '/lib/ai/AiGateway.php';
 
-        $this->ai = new AiGateway();
         $this->uploadVecs = [];
         $this->stepTimings = [];
 
@@ -141,7 +138,7 @@ final class DbnDeepResearchAgent
         if ($uploadChunks) {
             try {
                 $texts = array_map(fn(array $c) => $c['text'], $uploadChunks);
-                $vecs = $this->ai->embedBatch($texts, 'nomic-embed-text');
+                $vecs = dbnToolsLiteLLMEmbedBatch($texts);
                 if (count($vecs) === count($uploadChunks)) {
                     foreach ($uploadChunks as $i => $chunk) {
                         $this->uploadVecs[] = [
@@ -531,7 +528,7 @@ PROMPT;
             return [];
         }
         try {
-            $qVec = $this->ai->embed($question, 'nomic-embed-text');
+            $qVec = dbnToolsLiteLLMEmbedBatch([$question])[0] ?? [];
         } catch (Throwable $e) {
             error_log('DBN deep research sub-Q embed failed: ' . $e->getMessage());
             return [];

includes/bootstrap.php

@@ -676,3 +676,72 @@ function dbnToolsCallGpuLlm(array $messages, array $options = []): array
     }
     return $decoded;
 }
+
+/**
+ * Batch-embed texts via LiteLLM /v1/embeddings using cURL.
+ * Returns an array of float[] indexed by input position.
+ */
+function dbnToolsLiteLLMEmbedBatch(array $texts, string $model = 'nomic-embed-text', int $timeout = 60): array
+{
+    if (empty($texts)) {
+        return [];
+    }
+
+    $url    = 'http://10.0.1.10:4000/v1/embeddings';
+    $apiKey = (string)(dbnToolsEnv('LITELLM_MASTER_KEY') ?: 'sk-bnl-litellm-26xR9mK4qvN3wL8sTj7pB2d');
+
+    $payload = json_encode(['model' => $model, 'input' => array_values($texts)], JSON_UNESCAPED_UNICODE);
+    $headers = [
+        'Content-Type: application/json',
+        'Authorization: Bearer ' . $apiKey,
+    ];
+
+    if (function_exists('curl_init')) {
+        $ch = curl_init($url);
+        curl_setopt_array($ch, [
+            CURLOPT_RETURNTRANSFER => true,
+            CURLOPT_POST           => true,
+            CURLOPT_POSTFIELDS     => $payload,
+            CURLOPT_HTTPHEADER     => $headers,
+            CURLOPT_TIMEOUT        => $timeout,
+        ]);
+        $response = curl_exec($ch);
+        $code     = (int)curl_getinfo($ch, CURLINFO_RESPONSE_CODE);
+        $err      = curl_error($ch);
+        curl_close($ch);
+
+        if ($response === false) {
+            throw new RuntimeException('LiteLLM embed request failed: ' . $err);
+        }
+    } else {
+        $ctx = stream_context_create(['http' => [
+            'method'        => 'POST',
+            'header'        => implode("\r\n", $headers),
+            'content'       => $payload,
+            'timeout'       => $timeout,
+            'ignore_errors' => true,
+        ]]);
+        $response = @file_get_contents($url, false, $ctx);
+        $code     = 0;
+        if (isset($http_response_header[0]) && preg_match('/\s(\d{3})\s/', $http_response_header[0], $m)) {
+            $code = (int)$m[1];
+        }
+        if ($response === false) {
+            throw new RuntimeException('LiteLLM embed request failed.');
+        }
+    }
+
+    $decoded = json_decode($response, true);
+    if (!is_array($decoded)) {
+        throw new RuntimeException('LiteLLM embed returned non-JSON.');
+    }
+    if ($code < 200 || $code >= 300) {
+        $msg = $decoded['error']['message'] ?? ('HTTP ' . $code);
+        throw new RuntimeException('LiteLLM embed error: ' . $msg);
+    }
+
+    // OpenAI /v1/embeddings returns data[].embedding ordered by index
+    $data = $decoded['data'] ?? [];
+    usort($data, fn($a, $b) => ($a['index'] ?? 0) <=> ($b['index'] ?? 0));
+    return array_map(fn($d) => $d['embedding'], $data);
+}