diff --git a/README.md b/README.md
index 3f215cc..45e4423 100644
--- a/README.md
+++ b/README.md
@@ -11,6 +11,10 @@ To send a single query to an LLM, you can ask Neon to "ask Chat GPT <something>"
 To start conversing with an LLM, ask to "talk to Chat GPT" and have all of your input
 sent to an LLM until you say goodbye or stop talking for a while.
 
+Supported LLMs (subject to what your Neon Diana/Hana backend has deployed):
+Chat GPT, FastChat, Claude, and Gemini. Name any of them in a request,
+e.g. "ask Claude <something>" or "talk to Gemini".
+
 Enable fallback behavior by asking to "enable LLM fallback skill" or disable it
 by asking to "disable LLM fallback".
 
diff --git a/__init__.py b/__init__.py
index d640711..4374b6b 100644
--- a/__init__.py
+++ b/__init__.py
@@ -46,6 +46,8 @@
 class LLM(Enum):
     GPT = "Chat GPT"
     FASTCHAT = "FastChat"
+    CLAUDE = "Claude"
+    GEMINI = "Gemini"
 
 
 class LLMSkill(FallbackSkill):
@@ -53,7 +55,6 @@ def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
         self.chat_history = dict()
         self._default_user = "local"
-        self._default_llm = LLM.FASTCHAT
         self.chatting = dict()
         self.register_entity_file("llm.entity")
 
@@ -77,6 +78,23 @@ def chat_timeout_seconds(self):
     def fallback_enabled(self):
         return self.settings.get("fallback_enabled", False)
 
+    @property
+    def _default_llm(self) -> LLM:
+        """LLM used for the fallback handler and unrecognized requests.
+
+        Configurable via the `default_llm` setting (spoken name or enum name,
+        e.g. "chat gpt", "Chat GPT", or "GPT"); defaults to FastChat.
+        """
+        configured = self.settings.get("default_llm")
+        if not configured:
+            return LLM.FASTCHAT
+        requested = str(configured).strip().lower()
+        for llm in LLM:
+            if requested in (llm.value.lower(), llm.name.lower()):
+                return llm
+        LOG.warning(f"Unrecognized default_llm={configured!r}; using FastChat")
+        return LLM.FASTCHAT
+
     @fallback_handler(85)
     def fallback_llm(self, message):
         if not self.fallback_enabled:
@@ -86,7 +104,8 @@ def fallback_llm(self, message):
         LOG.info(f"Getting LLM response to: {utterance}")
         user = get_message_user(message) or self._default_user
 
-        def _threaded_get_response(utt, usr):
+        def _threaded_get_response(utt, usr, message):
+            # `message` required to resolve response routing in `speak`
             answer = self._get_llm_response(utt, usr, self._default_llm)
             if not answer:
                 LOG.info("No fallback response")
@@ -94,7 +113,8 @@ def _threaded_get_response(utt, usr):
             self.speak(answer)
 
         # TODO: Speak filler?
-        Thread(target=_threaded_get_response, args=(utterance, user), daemon=True).start()
+        Thread(target=_threaded_get_response,
+               args=(utterance, user, message), daemon=True).start()
         return True
 
     @intent_handler("enable_fallback.intent")
@@ -177,6 +197,10 @@ def _get_llm_response(self, query: str, user: str, llm: LLM) -> str:
             endpoint = "chatgpt"
         elif llm == LLM.FASTCHAT:
             endpoint = "fastchat"
+        elif llm == LLM.CLAUDE:
+            endpoint = "claude"
+        elif llm == LLM.GEMINI:
+            endpoint = "gemini"
         else:
             raise ValueError(f"Expected LLM, got: {llm}")
         self.chat_history.setdefault(user, list())
@@ -184,8 +208,9 @@ def _get_llm_response(self, query: str, user: str, llm: LLM) -> str:
 
         resp = resp.get("response") or ""
         if resp:
-            username = "user" if user == self._default_user else user
-            self.chat_history[user].append((username, query))
+            # History roles must be the literal "user"/"llm" the backend maps;
+            # the per-user key identifies the history, not the message role.
+            self.chat_history[user].append(("user", query))
             self.chat_history[user].append(("llm", resp))
         LOG.debug(f"Got LLM response: {resp}")
         return resp
@@ -196,6 +221,10 @@ def _get_requested_llm(self, message: Message) -> LLM:
             llm = LLM.GPT
         elif self.voc_match(request, "fastchat"):
             llm = LLM.FASTCHAT
+        elif self.voc_match(request, "claude"):
+            llm = LLM.CLAUDE
+        elif self.voc_match(request, "gemini"):
+            llm = LLM.GEMINI
         else:
             LOG.warning(f"No valid LLM in request: {request}")
             llm = LLM.GPT
diff --git a/locale/en-us/intent/ask_llm.intent b/locale/en-us/intent/ask_llm.intent
index 3d672c9..61e003f 100644
--- a/locale/en-us/intent/ask_llm.intent
+++ b/locale/en-us/intent/ask_llm.intent
@@ -1 +1 @@
-(ask|tell|task|half|sjet|asked|as) (chatgpt|chat gpt|chat g p t|chat gpg|chat gptc|jackgpt|jack gpt|jack chi pt|fast chat|fastchat|chat gbt|chatgbt|chat gpt4|chatgpt4|a language model|an llm) {question}
+(ask|tell|task|half|sjet|asked|as) (chatgpt|chat gpt|chat g p t|chat gpg|chat gptc|jackgpt|jack gpt|jack chi pt|fast chat|fastchat|chat gbt|chatgbt|chat gpt4|chatgpt4|claude|anthropic|gemini|bard|a language model|an llm) {question}
diff --git a/locale/en-us/intent/chat_with_llm.intent b/locale/en-us/intent/chat_with_llm.intent
index 1e35059..0793811 100644
--- a/locale/en-us/intent/chat_with_llm.intent
+++ b/locale/en-us/intent/chat_with_llm.intent
@@ -1 +1 @@
-(i want to |)(chat|talk|converse|start a conversation) (to|with) (chatgpt|chat gpt|chat g p t|chat gpg|chat gptc|jackgpt|jack gpt|jack chi pt|fast chat|fastchat|chat gbt|chatgbt|chat gpt4|chatgpt4|a language model|an llm)
+(i want to |)(chat|talk|converse|start a conversation) (to|with) (chatgpt|chat gpt|chat g p t|chat gpg|chat gptc|jackgpt|jack gpt|jack chi pt|fast chat|fastchat|chat gbt|chatgbt|chat gpt4|chatgpt4|claude|anthropic|gemini|bard|a language model|an llm)
diff --git a/locale/en-us/intent/llm.entity b/locale/en-us/intent/llm.entity
index 0b3b8cb..9748f58 100644
--- a/locale/en-us/intent/llm.entity
+++ b/locale/en-us/intent/llm.entity
@@ -10,3 +10,9 @@ chat gbt
 chatgbt
 chat gpt4
 chatgpt4
+fastchat
+fast chat
+claude
+anthropic
+gemini
+bard
diff --git a/locale/en-us/vocab/claude.voc b/locale/en-us/vocab/claude.voc
new file mode 100644
index 0000000..093c3e2
--- /dev/null
+++ b/locale/en-us/vocab/claude.voc
@@ -0,0 +1,3 @@
+claude
+claud
+anthropic
\ No newline at end of file
diff --git a/locale/en-us/vocab/gemini.voc b/locale/en-us/vocab/gemini.voc
new file mode 100644
index 0000000..16d8e27
--- /dev/null
+++ b/locale/en-us/vocab/gemini.voc
@@ -0,0 +1,5 @@
+gemini
+gemeni
+bard
+google's llm
+google llm
\ No newline at end of file
diff --git a/locale/uk-ua/intent/ask_llm.intent b/locale/uk-ua/intent/ask_llm.intent
index 5e6930c..a7fa114 100644
--- a/locale/uk-ua/intent/ask_llm.intent
+++ b/locale/uk-ua/intent/ask_llm.intent
@@ -1 +1 @@
-(запитай|скажи|попроси|задай питання|запитати|спитай|дізнайся у) (chatgpt|chat gpt|chat g p t|chat gpg|chat gptc|jackgpt|jack gpt|чат gpt|чатgpt|штучний інтелект|мовну модель|llm) {question}
+(запитай|скажи|попроси|задай питання|запитати|спитай|дізнайся у) (chatgpt|chat gpt|chat g p t|chat gpg|chat gptc|jackgpt|jack gpt|чат gpt|чатgpt|claude|клод|gemini|джеміні|bard|штучний інтелект|мовну модель|llm) {question}
diff --git a/locale/uk-ua/intent/chat_with_llm.intent b/locale/uk-ua/intent/chat_with_llm.intent
index 53456ca..e47a18a 100644
--- a/locale/uk-ua/intent/chat_with_llm.intent
+++ b/locale/uk-ua/intent/chat_with_llm.intent
@@ -1 +1 @@
-(я хочу |)(поспілкуватися|поговорити|почати розмову|спілкуватися) (з|із) (chatgpt|chat gpt|chat g p t|chat gpg|chat gptc|jackgpt|jack gpt|чат gpt|чатgpt|штучним інтелектом|мовною моделлю|llm)
+(я хочу |)(поспілкуватися|поговорити|почати розмову|спілкуватися) (з|із) (chatgpt|chat gpt|chat g p t|chat gpg|chat gptc|jackgpt|jack gpt|чат gpt|чатgpt|claude|клод|gemini|джеміні|bard|штучним інтелектом|мовною моделлю|llm)
diff --git a/locale/uk-ua/intent/llm.entity b/locale/uk-ua/intent/llm.entity
index 82ebe3f..7bfc5fc 100644
--- a/locale/uk-ua/intent/llm.entity
+++ b/locale/uk-ua/intent/llm.entity
@@ -14,3 +14,9 @@ chatgpt4
 чат gpt
 штучний інтелект
 мовна модель
+claude
+anthropic
+клод
+gemini
+bard
+джеміні
diff --git a/locale/uk-ua/vocab/claude.voc b/locale/uk-ua/vocab/claude.voc
new file mode 100644
index 0000000..fedf861
--- /dev/null
+++ b/locale/uk-ua/vocab/claude.voc
@@ -0,0 +1,5 @@
+claude
+anthropic
+клод
+клауд
+антропік
\ No newline at end of file
diff --git a/locale/uk-ua/vocab/gemini.voc b/locale/uk-ua/vocab/gemini.voc
new file mode 100644
index 0000000..a8cc8ac
--- /dev/null
+++ b/locale/uk-ua/vocab/gemini.voc
@@ -0,0 +1,5 @@
+gemini
+bard
+джеміні
+джемені
+бард
\ No newline at end of file
diff --git a/test/test_resources.yaml b/test/test_resources.yaml
index 4114424..ed98721 100644
--- a/test/test_resources.yaml
+++ b/test/test_resources.yaml
@@ -9,6 +9,8 @@ vocab:
   - exit
   - chat_gpt
   - fastchat
+  - claude
+  - gemini
 # dialog is .dialog file basenames (case-sensitive)
 dialog:
   - end_chat
diff --git a/test/test_skill.py b/test/test_skill.py
index ea5e0a4..a65357a 100644
--- a/test/test_skill.py
+++ b/test/test_skill.py
@@ -124,6 +124,43 @@ def test_handle_email_chat_history(self):
 
         self.skill._send_email = real_send_email
 
+    def test_get_requested_llm(self):
+        cases = {
+            "ask chat gpt something": LLM.GPT,
+            "talk to fastchat": LLM.FASTCHAT,
+            "ask claude something": LLM.CLAUDE,
+            "chat with gemini": LLM.GEMINI,
+        }
+        for utterance, expected in cases.items():
+            message = Message("test", {"utterance": utterance})
+            self.assertEqual(self.skill._get_requested_llm(message), expected,
+                             utterance)
+        # Unrecognized LLM falls back to GPT
+        message = Message("test", {"utterance": "ask the robot something"})
+        self.assertEqual(self.skill._get_requested_llm(message), LLM.GPT)
+
+    def test_get_llm_response_endpoints(self):
+        from unittest.mock import patch
+        expected = {
+            LLM.GPT: "chatgpt",
+            LLM.FASTCHAT: "fastchat",
+            LLM.CLAUDE: "claude",
+            LLM.GEMINI: "gemini",
+        }
+        real_history = self.skill.chat_history
+        try:
+            for llm, endpoint in expected.items():
+                self.skill.chat_history = dict()
+                with patch("skill_fallback_llm.request_backend",
+                           return_value={"response": "hi"}) as mock_backend:
+                    resp = self.skill._get_llm_response("hello", "endpoint_user",
+                                                        llm)
+                self.assertEqual(resp, "hi")
+                self.assertEqual(mock_backend.call_args[0][0],
+                                 f"/llm/{endpoint}")
+        finally:
+            self.skill.chat_history = real_history
+
     def test_converse(self):
         # TODO
         pass