feat(generate): add memory to openai

src/cbrkit/generate/__init__.py

@@ -1,5 +1,5 @@
 import asyncio
-from collections.abc import Callable, Sequence
+from collections.abc import Callable, MutableSequence, Sequence
 from dataclasses import dataclass, field
 from typing import cast
 
@@ -58,11 +58,15 @@ def __call__(self, prompts: Sequence[str]) -> Sequence[T]:
     from openai import AsyncOpenAI, pydantic_function_tool
     from openai.types.chat import ChatCompletionMessageParam
 
-    @dataclass(slots=True, frozen=True)
+    @dataclass(slots=True)
     class openai[T: BaseModel | str](GenerationSeqFunc[T]):
         model: str
         schema: type[T]
-        messages: Sequence[ChatCompletionMessageParam] = field(default_factory=tuple)
+        messages: MutableSequence[ChatCompletionMessageParam] = field(
+            default_factory=list
+        )
+        memorize: bool = False
+        memorize_func: Callable[[T], str] = str
         client: AsyncOpenAI = field(default_factory=AsyncOpenAI, repr=False)
 
         def __call__(self, prompts: Sequence[str]) -> Sequence[T]:
@@ -85,6 +89,8 @@ async def _generate_single(self, prompt: str) -> T:
                 },
             ]
 
+            result: T | None = None
+
             if self.schema is BaseModel:
                 tool = pydantic_function_tool(cast(type[BaseModel], self.schema))
 
@@ -108,19 +114,28 @@ async def _generate_single(self, prompt: str) -> T:
                 if parsed is None:
                     raise ValueError("The tool call is empty")
 
-                return cast(T, parsed)
+                result = cast(T, parsed)
 
-            res = await self.client.beta.chat.completions.parse(
-                model=self.model,
-                messages=messages,
-            )
+            else:
+                res = await self.client.beta.chat.completions.parse(
+                    model=self.model,
+                    messages=messages,
+                )
+
+                content = res.choices[0].message.content
 
-            content = res.choices[0].message.content
+                if content is None:
+                    raise ValueError("The completion is empty")
 
-            if content is None:
-                raise ValueError("The completion is empty")
+                result = cast(T, content)
+
+            if self.memorize:
+                self.messages.append({"role": "user", "content": prompt})
+                self.messages.append(
+                    {"role": "system", "content": self.memorize_func(result)}
+                )
 
-            return cast(T, content)
+            return result
 
     __all__ += ["openai"]