diff --git a/chatmastermind/ai.py b/chatmastermind/ai.py
new file mode 100644
index 0000000..4a8b914
--- /dev/null
+++ b/chatmastermind/ai.py
@@ -0,0 +1,63 @@
+from dataclasses import dataclass
+from typing import Protocol, Optional, Union
+from .configuration import AIConfig
+from .tags import Tag
+from .message import Message
+from .chat import Chat
+
+
+class AIError(Exception):
+    pass
+
+
+@dataclass
+class Tokens:
+    prompt: int = 0
+    completion: int = 0
+    total: int = 0
+
+
+@dataclass
+class AIResponse:
+    """
+    The response to an AI request. Consists of one or more messages
+    (each containing the question and a single answer) and the nr.
+    of used tokens.
+    """
+    messages: list[Message]
+    tokens: Optional[Tokens] = None
+
+
+class AI(Protocol):
+    """
+    The base class for AI clients.
+    """
+
+    name: str
+    config: AIConfig
+
+    def request(self,
+                question: Message,
+                context: Chat,
+                num_answers: int = 1,
+                otags: Optional[set[Tag]] = None) -> AIResponse:
+        """
+        Make an AI request, asking the given question with the given
+        context (i. e. chat history). The nr. of requested answers
+        corresponds to the nr. of messages in the 'AIResponse'.
+        """
+        raise NotImplementedError
+
+    def models(self) -> list[str]:
+        """
+        Return all models supported by this AI.
+        """
+        raise NotImplementedError
+
+    def tokens(self, data: Union[Message, Chat]) -> int:
+        """
+        Computes the nr. of AI language tokens for the given message
+        or chat. Note that the computation may not be 100% accurate
+        and is not implemented for all AIs.
+        """
+        raise NotImplementedError