diff --git a/scripts/native_client.py b/scripts/native_client.py
index b869b757..2f26f713 100755
--- a/scripts/native_client.py
+++ b/scripts/native_client.py
@@ -117,6 +117,14 @@ async def translate(self, text, src=None, trg=None, *, model=None, pivot=None, h
     async def download_model(self, model_id, *, update=lambda data: None):
         return await self.request("DownloadModel", {"modelID": str(model_id)}, update=update)
 
+    async def configure(self, *, threads:int = None, cache_size:int = None):
+        options = {}
+        if threads is not None:
+            options["threads"] = int(threads)
+        if cache_size is not None:
+            options["cacheSize"] = int(cache_size)
+        return await self.request("Configure", options)
+
 
 def first(iterable, *default):
     """Returns the first value of anything iterable, or throws StopIteration
@@ -186,6 +194,8 @@ async def test():
             if model["id"] == selected_model["id"]
         )
 
+        await tl.configure(threads=1, cache_size=0)
+
         # Perform some translations, switching between the models
         translations = await asyncio.gather(
             tl.translate("Hello world!", "en", "de"),
diff --git a/src/cli/NativeMsgIface.cpp b/src/cli/NativeMsgIface.cpp
index 8dbd6726..47d4042a 100644
--- a/src/cli/NativeMsgIface.cpp
+++ b/src/cli/NativeMsgIface.cpp
@@ -224,6 +224,17 @@ void NativeMsgIface::handleRequest(ListRequest request)  {
     writeResponse(request, modelsJson);
 }
 
+void NativeMsgIface::handleRequest(ConfigureRequest request)  {
+    marian::bergamot::AsyncService::Config serviceConfig;
+    serviceConfig.numWorkers = request.threads;
+    serviceConfig.cacheSize = request.cacheSize;
+    service_.reset();
+    service_ = std::make_shared<marian::bergamot::AsyncService>(serviceConfig);
+    
+    QJsonObject response{}; // I don't know...
+    writeResponse(request, response);
+}
+
 void NativeMsgIface::handleRequest(DownloadRequest request)  {
     // Edge case: client issued a DownloadRequest before fetching the list of
     // remote models because it knows the model ID from a previous run. We still
@@ -284,7 +295,7 @@ request_variant NativeMsgIface::parseJsonInput(QByteArray input) {
 
     // Define what are mandatory and what are optional request keys
     static const QStringList mandatoryKeys({"command", "id", "data"}); // Expected in every message
-    static const QSet<QString> commandTypes({"ListModels", "DownloadModel", "Translate"});
+    static const QSet<QString> commandTypes({"ListModels", "DownloadModel", "Translate", "Configure"});
     // Json doesn't have schema validation, so validate here, in place:
     QString command;
     int id;
@@ -359,13 +370,21 @@ request_variant NativeMsgIface::parseJsonInput(QByteArray input) {
         ret.id = id;
         for (auto&& key : mandatoryKeysDownload) {
             QJsonValueRef val = data[key];
-            if (val.isNull()) {
+            if (val.isNull() || val.isUndefined()) {
                 return MalformedRequest{id, QString("data field key %1 cannot be null!").arg(key)};
             } else {
                 ret.modelID = val.toString();
             }
         }
         return ret;
+    } else if (command == "Configure") {
+        ConfigureRequest ret;
+        ret.id = id;
+        if (!data["threads"].isUndefined())
+            ret.threads = data["threads"].toInt();
+        if (!data["cacheSize"].isUndefined())
+            ret.cacheSize = data["cacheSize"].toInt();
+        return ret;
     } else {
         return MalformedRequest{id, QString("Developer error. We shouldn't ever be here! Command: %1").arg(command)};
     }
diff --git a/src/cli/NativeMsgIface.h b/src/cli/NativeMsgIface.h
index 6bad9aa0..88071d3f 100644
--- a/src/cli/NativeMsgIface.h
+++ b/src/cli/NativeMsgIface.h
@@ -236,6 +236,33 @@ struct DownloadRequest : Request {
 
 Q_DECLARE_METATYPE(DownloadRequest);
 
+/**
+ * Change TranslateLocally resource usage for this session.
+ * 
+ * Request:
+ * {
+ *   "id": int,
+ *   "command": "Configure",
+ *   "data": {
+ *     "threads": int
+ *     "cacheSize": int (0 means disabled)
+ *   }
+ * }
+ * 
+ * Successful response:
+ * {
+ *   "id": int,
+ *   "success": true,
+ *   "data": {}
+ * }
+ */
+struct ConfigureRequest : Request {
+    int threads;
+    int cacheSize;
+};
+
+Q_DECLARE_METATYPE(ConfigureRequest);
+
 /**
  * Internal structure to handle a request that is missing a required field.
  */
@@ -243,7 +270,7 @@ struct MalformedRequest : Request {
     QString error;
 };
 
-using request_variant = std::variant<TranslationRequest, ListRequest, DownloadRequest, MalformedRequest>;
+using request_variant = std::variant<TranslationRequest, ListRequest, DownloadRequest, ConfigureRequest, MalformedRequest>;
 
 /**
  * Internal structure to cache a loaded direct model (i.e. no pivoting)
@@ -410,6 +437,10 @@ private slots:
      */
     void handleRequest(DownloadRequest myJsonInput);
 
+    /**
+     */
+    void handleRequest(ConfigureRequest myJsonInput);
+
     /**
      * @brief handleRequest handles a request type MalformedRequest and writes to stdout
      * @param myJsonInput MalformedRequest