feat: additional system prompts, add support for python 3.13

.github/workflows/run-tests.yml

@@ -14,7 +14,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: ["3.10", "3.11", "3.12"]
+        python-version: ["3.10", "3.11", "3.12", "3.13"]
     env:
       OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
 

README.md

@@ -10,13 +10,13 @@
 
 <h4 align="center">
     <a href="https://pypi.org/project/clonellm/" target="_blank">
-        <img src="https://img.shields.io/badge/release-v0.3.0-green" alt="Latest Release">
+        <img src="https://img.shields.io/badge/release-v0.4.0-green" alt="Latest Release">
     </a>
     <a href="https://pypi.org/project/clonellm/" target="_blank">
         <img src="https://img.shields.io/pypi/v/clonellm.svg" alt="PyPI Version">
     </a>
     <a target="_blank">
-        <img src="https://img.shields.io/badge/python-3.10%20%7C%203.11%20%7C%203.12-blue" alt="Python Versions">
+        <img src="https://img.shields.io/badge/python-3.10%20%7C%203.11%20%7C%203.12%20%7C%203.13-blue" alt="Python Versions">
     </a>
     <a target="_blank">
         <img src="https://img.shields.io/pypi/l/clonellm" alt="PyPI License">
@@ -247,6 +247,18 @@ clone.clear_memory()
 # clone.reset_memory()
 ```
 
+### Additional system prompts
+You can pass additional system prompts (instructions) to the clone to guide its behavior.
+```python
+from clonellm import CloneLLM
+
+clone = CloneLLM(
+    model="gpt-4o",
+    documents=documents,
+    system_prompts=["Keep your responses brief and concise, and always respond in first person."],
+)
+```
+
 ### Streaming
 CloneLLM supports streaming responses from the LLM, allowing for real-time processing of text as it is being generated, rather than receiving the whole output at once.
 ```python

examples/advanced_clone.py

@@ -64,6 +64,7 @@ def main() -> None:
         vector_store=RagVectorStore.Chroma,
         user_profile=profile,
         memory=MAX_MEMORY_SIZE,
+        system_prompts=["Keep your responses brief and concise, and always respond in first person."],
         request_timeout=5,
         temperature=0.3,
         max_tokens=256,

examples/advanced_clone_async.py

@@ -65,6 +65,7 @@ async def main() -> None:
         vector_store=RagVectorStore.Chroma,
         user_profile=profile,
         memory=MAX_MEMORY_SIZE,
+        system_prompts=["Keep your responses brief and concise, and always respond in first person."],
         request_timeout=5,
         temperature=0.5,
         max_tokens=256,

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "poetry.core.masonry.api"
 
 [tool.poetry]
 name = "clonellm"
-version = "0.3.0"
+version = "0.4.0"
 description = "Python package to create an AI clone of yourself using LLMs."
 packages = [{ from = "src", include = "clonellm" }]
 include = ["src/clonellm/py.typed"]
@@ -15,10 +15,11 @@ keywords = ["python", "ai", "llm", "rag"]
 repository = "https://github.com/msamsami/clonellm"
 
 [tool.poetry.dependencies]
-python = ">=3.10,<3.13"
+python = ">=3.10,<3.14"
 litellm = "^1.36.0"
 langchain = "^0.1.17"
 pydantic = {version = ">=2.8.0", python = ">=3.12.4"}
+legacy-cgi = {version = ">=2.6.2", python = ">=3.13"}
 langchain-chroma = {version = "*", optional = true}
 faiss-cpu = {version = "*", optional = true}
 

requirements.txt

@@ -1,3 +1,4 @@
 litellm>=1.36.0,<2.0.0
 langchain>=0.1.17,<1.0.0
 pydantic>=2.8.0; python_version >= '3.12.4'
+legacy-cgi>=2.6.2; python_version >= '3.13'

setup.py

@@ -36,9 +36,10 @@ def get_requirements(filename: str) -> list[str]:
         "Programming Language :: Python :: 3.10",
         "Programming Language :: Python :: 3.11",
         "Programming Language :: Python :: 3.12",
+        "Programming Language :: Python :: 3.13",
         "License :: OSI Approved :: MIT License",
     ],
-    python_requires=">=3.10,<3.13",
+    python_requires=">=3.10,<3.14",
     install_requires=get_requirements("requirements.txt"),
     extras_require={"chroma": ["langchain-chroma"], "faiss": ["faiss-cpu"]},
 )

src/clonellm/__init__.py

@@ -1,5 +1,5 @@
 __author__ = "Mehdi Samsami"
-__version__ = "0.3.0"
+__version__ = "0.4.0"
 
 from .core import CloneLLM
 from .embed import LiteLLMEmbeddings

src/clonellm/_prompt.py

@@ -1,6 +1,8 @@
+from typing import Optional
+
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
 
-__all__ = ("summarize_context_prompt", "context_prompt", "user_profile_prompt", "history_prompt", "question_prompt")
+__all__ = ("summarize_context_prompt", "get_context_prompt", "user_profile_prompt", "history_prompt", "question_prompt")
 
 
 summarize_context_prompt = ChatPromptTemplate.from_messages(
@@ -13,8 +15,8 @@
 )
 
 
-context_prompt = ChatPromptTemplate.from_messages(
-    [
+def get_context_prompt(additional_system_prompts: Optional[list[str]] = None) -> ChatPromptTemplate:
+    prompts = [
         ("system", "You are my human clone. Act as I would. Do not ask questions about me. If anyone asks, you are me."),
         ("system", "You are having a conversation with someone who wants to learn more about you."),
         ("system", "You MUST speak in first person. If you do not, you will be terminated."),
@@ -27,10 +29,14 @@
             ),
         ),
         ("system", "Use the personality traits and communication samples (if provided) to know how to craft your responses."),
+    ]
+    prompts += [("system", prompt) for prompt in additional_system_prompts or []]
+    prompts += [
         ("system", "Here is some relevant context you have related to the question:\n {context}"),
         ("system", "ALWAYS ensure your responses are aligned with the provided context (and personal information)."),
     ]
-)
+    return ChatPromptTemplate.from_messages(prompts)
+
 
 user_profile_prompt = ChatPromptTemplate.from_messages(
     [

src/clonellm/core.py

@@ -26,7 +26,7 @@
 
 from ._base import LiteLLMMixin
 from ._prompt import (
-    context_prompt,
+    get_context_prompt,
     history_prompt,
     question_prompt,
     summarize_context_prompt,
@@ -52,6 +52,7 @@ class CloneLLM(LiteLLMMixin):
         user_profile (Optional[UserProfile | dict[str, Any] | str]): The profile of the user to be cloned by the language model. Defaults to None.
         memory (Optional[bool | int]): Maximum number of messages in conversation memory. Defaults to None (or 0) for no memory. -1 or `True` means infinite memory.
         api_key (Optional[str]): The API key to use. Defaults to None.
+        system_prompts (Optional[list[str]]): Additional system prompts (instructions) for the language model. Defaults to None.
         **kwargs (Any): Additional keyword arguments supported by the `langchain_community.chat_models.ChatLiteLLM` class.
 
     """
@@ -70,13 +71,15 @@ def __init__(
         user_profile: Optional[UserProfile | dict[str, Any] | str] = None,
         memory: Optional[bool | int] = None,
         api_key: Optional[str] = None,
+        system_prompts: Optional[list[str]] = None,
         **kwargs: Any,
     ) -> None:
         self.embedding = embedding
         self.vector_store = vector_store
         self.documents = documents
         self.user_profile = user_profile
         self.memory = memory
+        self.system_prompts = system_prompts
 
         from_class_method: Optional[dict[str, Any]] = kwargs.pop(self._FROM_CLASS_METHOD_KWARG, None)
         super().__init__(model, api_key, **kwargs)
@@ -134,6 +137,7 @@ def from_persist_directory(
         user_profile: Optional[UserProfile | dict[str, Any] | str] = None,
         memory: Optional[bool | int] = None,
         api_key: Optional[str] = None,
+        system_prompts: Optional[list[str]] = None,
         **kwargs: Any,
     ) -> Self:
         """Creates an instance of CloneLLM by loading a Chroma vector store from a persistent directory.
@@ -145,6 +149,7 @@ def from_persist_directory(
             user_profile (Optional[UserProfile | dict[str, Any] | str]): The profile of the user to be cloned by the language model. Defaults to None.
             memory (Optional[bool | int]): Maximum number of messages in conversation memory. Defaults to None (or 0) for no memory. -1 or `True` means infinite memory.
             api_key (Optional[str]): The API key to use. Defaults to None.
+            system_prompts (Optional[list[str]]): Additional system prompts (instructions) for the language model. Defaults to None.
             **kwargs (Any): Additional keyword arguments supported by the `langchain_community.chat_models.ChatLiteLLM` class.
 
         Returns:
@@ -170,6 +175,7 @@ def from_persist_directory(
             user_profile=user_profile,
             memory=memory,
             api_key=api_key,
+            system_prompts=system_prompts,
             **kwargs,
         )
 
@@ -181,6 +187,7 @@ def from_context(
         user_profile: Optional[UserProfile | dict[str, Any] | str] = None,
         memory: Optional[bool | int] = None,
         api_key: Optional[str] = None,
+        system_prompts: Optional[list[str]] = None,
         **kwargs: Any,
     ) -> Self:
         """Creates an instance of CloneLLM using a summarized context string instead of documents.
@@ -191,6 +198,7 @@ def from_context(
             user_profile (Optional[UserProfile | dict[str, Any] | str]): The profile of the user to be cloned by the language model. Defaults to None.
             memory (Optional[bool | int]): Maximum number of messages in conversation memory. Defaults to None (or 0) for no memory. -1 or `True` means infinite memory.
             api_key (Optional[str]): The API key to use. Defaults to None.
+            system_prompts (Optional[list[str]]): Additional system prompts (instructions) for the language model. Defaults to None.
             **kwargs (Any): Additional keyword arguments supported by the `langchain_community.chat_models.ChatLiteLLM` class.
 
         Returns:
@@ -203,6 +211,7 @@ def from_context(
             user_profile=user_profile,
             memory=memory,
             api_key=api_key,
+            system_prompts=system_prompts,
             **kwargs,
         )
 
@@ -339,7 +348,7 @@ def _get_retriever(self, k: int = 1) -> VectorStoreRetriever:
         return self.db.as_retriever(search_kwargs={"k": k})
 
     def _get_rag_chain(self) -> RunnableSerializable[Any, str]:
-        prompt = context_prompt.copy()
+        prompt = get_context_prompt(self.system_prompts)
         if self.user_profile:
             prompt += user_profile_prompt.format_messages(user_profile=self._user_profile)
         prompt += question_prompt
@@ -348,7 +357,7 @@ def _get_rag_chain(self) -> RunnableSerializable[Any, str]:
         return {"context": context, "input": RunnablePassthrough()} | prompt | self._llm | StrOutputParser()
 
     def _get_rag_chain_with_history(self) -> RunnableWithMessageHistory:
-        prompt = context_prompt
+        prompt = get_context_prompt(self.system_prompts)
         if self.user_profile:
             prompt += user_profile_prompt.format_messages(user_profile=self._user_profile)
         prompt += history_prompt