diff --git a/src/Core/RequestHandler.php b/src/Core/RequestHandler.php
index 4a022aa5..2a9689c8 100644
--- a/src/Core/RequestHandler.php
+++ b/src/Core/RequestHandler.php
@@ -6,7 +6,6 @@
 use Cognesy\Instructor\Data\Request;
 use Cognesy\Instructor\Enums\Mode;
 use Cognesy\Instructor\Events\EventDispatcher;
-use Cognesy\Instructor\Events\Instructor\InstructorDone;
 use Cognesy\Instructor\Events\Instructor\ResponseGenerated;
 use Cognesy\Instructor\Events\Request\NewValidationRecoveryAttempt;
 use Cognesy\Instructor\Events\Request\ValidationRecoveryLimitReached;
@@ -16,7 +15,6 @@
 use Cognesy\Instructor\Extras\LLM\Data\PartialLLMResponse;
 use Cognesy\Instructor\Extras\LLM\Inference;
 use Cognesy\Instructor\Extras\LLM\InferenceResponse;
-use Cognesy\Instructor\Stream;
 use Cognesy\Instructor\Utils\Json\Json;
 use Cognesy\Instructor\Utils\Result\Result;
 use Exception;
@@ -39,40 +37,16 @@ public function __construct(
         EventDispatcher $events,
     ) {
         $this->events = $events;
+        $this->retries = 0;
+        $this->errors = [];
     }
 
-    /**
-     * Executes the request and returns the response
-     */
-    public function get() : mixed {
-        if ($this->request->isStream()) {
-            return $this->stream()->final();
-        }
-        $result = $this->responseFor($this->request);
-        $this->events->dispatch(new InstructorDone(['result' => $result]));
-        return $result->value();
-    }
-
-    /**
-     * Executes the request and returns the response stream
-     */
-    public function stream() : Stream {
-        // TODO: do we need this? cannot we just turn streaming on?
-        if (!$this->request->isStream()) {
-            throw new Exception('Instructor::stream() method requires response streaming: set "stream" = true in the request options.');
-        }
-        $stream = $this->streamResponseFor($this->request);
-        return new Stream($stream, $this->events);
-    }
-
-    // INTERNAL //////////////////////////////////////////////////////////////
+    // PUBLIC //////////////////////////////////////////////////////////////
 
     /**
      * Generates response value
      */
-    protected function responseFor(Request $request) : LLMResponse {
-        $this->init();
-
+    public function responseFor(Request $request) : LLMResponse {
         $processingResult = Result::failure("No response generated");
         while ($processingResult->isFailure() && !$this->maxRetriesReached($request)) {
             $llmResponse = $this->getInference($request)->toLLMResponse();
@@ -94,9 +68,7 @@ protected function responseFor(Request $request) : LLMResponse {
      * @param Request $request
      * @return Generator<PartialLLMResponse|LLMResponse>
      */
-    protected function streamResponseFor(Request $request) : Generator {
-        $this->init();
-
+    public function streamResponseFor(Request $request) : Generator {
         $processingResult = Result::failure("No response generated");
         while ($processingResult->isFailure() && !$this->maxRetriesReached($request)) {
             $stream = $this->getInference($request)->toPartialLLMResponses();
@@ -112,10 +84,7 @@ protected function streamResponseFor(Request $request) : Generator {
         yield $llmResponse->withValue($value);
     }
 
-    protected function init() : void {
-        $this->retries = 0;
-        $this->errors = [];
-    }
+    // INTERNAL ///////////////////////////////////////////////////////////
 
     protected function getInference(Request $request) : InferenceResponse {
         $inference = new Inference(
diff --git a/src/InstructorResponse.php b/src/InstructorResponse.php
new file mode 100644
index 00000000..5974830d
--- /dev/null
+++ b/src/InstructorResponse.php
@@ -0,0 +1,50 @@
+<?php
+
+namespace Cognesy\Instructor;
+
+use Cognesy\Instructor\Core\RequestHandler;
+use Cognesy\Instructor\Data\Request;
+use Cognesy\Instructor\Events\EventDispatcher;
+use Cognesy\Instructor\Events\Instructor\InstructorDone;
+use Exception;
+
+class InstructorResponse
+{
+    private RequestHandler $requestHandler;
+    private EventDispatcher $events;
+    private Request $request;
+
+    public function __construct(
+        Request $request,
+        RequestHandler $requestHandler,
+        EventDispatcher $events,
+    ) {
+        $this->events = $events;
+        $this->requestHandler = $requestHandler;
+        $this->request = $request;
+    }
+
+    /**
+     * Executes the request and returns the response
+     */
+    public function get() : mixed {
+        if ($this->request->isStream()) {
+            return $this->stream()->final();
+        }
+        $result = $this->requestHandler->responseFor($this->request);
+        $this->events->dispatch(new InstructorDone(['result' => $result]));
+        return $result->value();
+    }
+
+    /**
+     * Executes the request and returns the response stream
+     */
+    public function stream() : Stream {
+        // TODO: do we need this? cannot we just turn streaming on?
+        if (!$this->request->isStream()) {
+            throw new Exception('Instructor::stream() method requires response streaming: set "stream" = true in the request options.');
+        }
+        $stream = $this->requestHandler->streamResponseFor($this->request);
+        return new Stream($stream, $this->events);
+    }
+}
\ No newline at end of file
diff --git a/src/Traits/Instructor/HandlesInvocation.php b/src/Traits/Instructor/HandlesInvocation.php
index cc36d31b..ae489dd7 100644
--- a/src/Traits/Instructor/HandlesInvocation.php
+++ b/src/Traits/Instructor/HandlesInvocation.php
@@ -9,6 +9,9 @@
 use Cognesy\Instructor\Data\ResponseModel;
 use Cognesy\Instructor\Enums\Mode;
 use Cognesy\Instructor\Events\Instructor\RequestReceived;
+use Cognesy\Instructor\Extras\LLM\Inference;
+use Cognesy\Instructor\Extras\LLM\InferenceResponse;
+use Cognesy\Instructor\InstructorResponse;
 use Cognesy\Instructor\Schema\Factories\SchemaFactory;
 use Cognesy\Instructor\Schema\Factories\ToolCallBuilder;
 use Cognesy\Instructor\Schema\Utils\ReferenceQueue;
@@ -69,7 +72,7 @@ public function request(
         string              $toolDescription = '',
         string              $retryPrompt = '',
         Mode                $mode = Mode::Tools,
-    ) : RequestHandler {
+    ) : InstructorResponse {
         $this->queueEvent(new RequestReceived());
         $this->dispatchQueuedEvents();
 
@@ -98,7 +101,7 @@ public function request(
             cachedContext: $this->cachedContext ?? [],
         );
 
-        return new RequestHandler(
+        $requestHandler = new RequestHandler(
             request: $request,
             responseGenerator: new ResponseGenerator(
                 $this->responseDeserializer,
@@ -116,10 +119,35 @@ public function request(
             httpClient: $this->httpClient,
             events: $this->events,
         );
+
+        return new InstructorResponse(
+            request: $request,
+            requestHandler: $requestHandler,
+            events: $this->events,
+        );
     }
 
     // INTERNAL /////////////////////////////////////////////////
 
+    protected function getInference(Request $request) : InferenceResponse {
+        $inference = new Inference(
+            connection: $this->connection,
+            httpClient: $this->httpClient,
+            driver: $this->driver,
+            events: $this->events,
+        );
+        return $inference
+            ->create(
+                $request->toMessages(),
+                $request->model(),
+                $request->toolCallSchema(),
+                $request->toolChoice(),
+                $request->responseFormat(),
+                $request->options(),
+                $request->mode()
+            );
+    }
+
     private function makeResponseModel(
         string|array|object $requestedSchema,
         string $toolName,