added publishing and improved rag example

.pypirc

@@ -0,0 +1,5 @@
+[distutils]
+index-servers=pypi
+[pypi]
+repository=https://upload.pypi.org/legacy/
+username=antonkulaga

README.md

@@ -65,4 +65,4 @@ for _ in range(exchanges):
 
 All prompts that we use are stored in yaml files that you can easily overload.
 
-The only complex dependency that we use is Mako for prompt templates.
+The only complex (but not mandatory) dependency that we use is Mako for prompt templates

environment.yaml

@@ -6,10 +6,10 @@ dependencies:
   - python=3.10
   - pip
   - loguru
-  - click
   - requests
-  - pytest
-  - typer
+  - pytest #for tests
+  - typer #for CLI
+  - twine #for publishing
   - pip:
       - litellm>=1.40.6
       - jupyter

examples/rag.py

@@ -1,4 +1,5 @@
 import pprint
+import sys
 from pathlib import Path
 from typing import Dict, Any
 
@@ -15,6 +16,11 @@
 
 app = typer.Typer(no_args_is_help=True)
 
+def configure_logger(level: str) -> None:
+    """Configure the logger to the specified log level."""
+    logger.remove()
+    logger.add(sys.stdout, level=level)
+
 @app.command()
 def search(query: str, limit: int = 10):
     """
@@ -31,35 +37,50 @@ def search(query: str, limit: int = 10):
 
 
 @app.command()
-def rapamycin():
+def rapamycin(prompt_name: str = "rapamycin_case", sub_prompt: str = "with_requirements", log_level: str = "INFO"):
+    configure_logger(log_level)
     logger.add("logs/rag_rapamycin.txt", rotation="1 MB")
     load_dotenv()
+
+
+    # setting up relative paths to define output and load prompts
+    current_folder: Path = Path(__file__).parent
+    example_prompts = current_folder / "example_prompts.yaml"
+    output = Path(__file__).parent.parent / "output" / "examples"
+    prompts = yaml.safe_load(example_prompts.open("r"))
+    question = prompts[prompt_name][sub_prompt]
+
+
     scientist: ChatAgent = ChatAgent(llm_options = LLAMA3,
                                     role = "scientist",
                                     goal = "Research the topics in the most comprehensive way, using search in academic literature and providing sources",
                                     task="Address the research question in the most comprehensive way",
                                     tools = [literature_search])
 
+    # ADDING LOGGER HANDLERS:
+    scientist.memory.add_on_message(lambda m: logger.debug(f"SCIENTIST MESSAGE: {m}"))
+    scientist.memory.add_on_tool_call(lambda f: logger.debug(f"SCIENTIST FUNCTION: {f}"))
+    scientist.memory.add_on_tool_result(lambda m: logger.debug(f"SCIENTIST TOOL result from {m.name} with tool call id {m.tool_call_id} is {m.content}"))
+
+
     critic: ChatAgent = ChatAgent(llm_options = LLAMA3,
-                                  role = "critic",
-                                  goal = "Evaluate the answer according to the criteria provided",
-                                  task="Evaluate the answer according to the criteria provided and make recommendations to improve")
-    #TODO: write down cryticism
+                              role = "critic",
+                              goal = "Evaluate the answer according to the criteria provided",
+                              task="Evaluate the answer according to the criteria provided and make recommendations to improve")
 
-    scientist.memory.add_on_message(lambda m: logger.info(f"MESSAGE: {m}"))
-    scientist.memory.add_on_tool_call(lambda t: logger.info(f"FUNCTION: {t}"))
+    # ADDING CRITICS HANDLERS:
+    critic.memory.add_on_message(lambda m: logger.debug(f"CRITIC MESSAGE: {m}"))
 
-    # setting up relative paths to define output and load prompts
-    current_folder: Path = Path(__file__).parent
-    example_prompts = current_folder / "example_prompts.yaml"
-    output = Path(__file__).parent.parent / "output" / "examples"
-    prompts = yaml.safe_load(example_prompts.open("r"))
+    answer = scientist.query(question, output=output / prompt_name / f"{sub_prompt}_initial_answer.txt")
+    logger.info(f"INITIAL ANSWER: {answer}")
 
-    rapamycin = prompts["rapamycin_case"]["with_requirements"]
+    for_review = f"""
+    The question that scientist asked was: {question}
+    The answer that she gave was: {answer}
+    """
 
-    result = scientist.query(rapamycin, output=output / "rapamycin" / "with_requirements.txt")
-    print("RESULT IS:")
-    pprint.pprint(result)
+    review_results = critic.query(for_review, output=output / prompt_name / f"{sub_prompt}_answer_review.txt")
+    logger.info(f"REVIEW RESULTS: {review_results}")
 
 
 

examples/two_agents_chat.py

@@ -23,7 +23,6 @@ def purchase_example():
 
     exchanges: int = 3
     customer.memory.add_on_message(lambda m: logger.info(f"Customer: {m}") if m.role == "user" else logger.info(f"Storekeeper: {m}"))
-
     customer_reply = "Hi."
     for _ in range(exchanges):
         storekeeper_reply = storekeeper.query(customer_reply)

just_agents/llm_session.py

@@ -191,6 +191,8 @@ def _query(self, run_callbacks: bool = True, output: Optional[Path] = None) -> s
         result: str = self.memory.last_message.content if self.memory.last_message is not None and self.memory.last_message.content is not None else str(
             self.memory.last_message)
         if output is not None:
+            if not output.parent.exists():
+                output.parent.mkdir(parents=True, exist_ok=True)
             output.write_text(result)
         return result
 

just_agents/memory.py

@@ -16,6 +16,9 @@ class Memory:
     def add_on_message(self, handler: OnMessageCallable):
         self.on_message.append(handler)
 
+    def add_on_tool_result(self, handler: OnMessageCallable):
+        self.add_on_message(lambda m: handler(m) if m.role == "tool" else None)
+
     def add_on_tool_call(self, fun: OnFunctionCallable):
         """
         Adds handler only to function calls to track what exactly was called
@@ -34,6 +37,12 @@ def tool_handler(message: Message) -> None:
     def remove_on_message(self, handler: OnMessageCallable):
         self.on_message = [m for m in self.on_message if m == handler]
 
+    def add_system_message(self, prompt: str, run_callbacks: bool = True):
+        return self.add_message(Message(role="system", content=prompt), run_callbacks=run_callbacks)
+
+    def add_user_message(self, prompt: str, run_callbacks: bool = True):
+        return self.add_message(Message(role="user", content=prompt), run_callbacks=run_callbacks)
+
 
     def add_message(self, message: Message, run_callbacks: bool = True):
         """

just_agents/tools/search.py

@@ -67,7 +67,6 @@ def get_semantic_paper(query: str,
     return results
 
 
-
 def hybrid_search(text: str,
                   collections = ["aging_papers_paragraphs_bge_base_en_v1.5", "aging_papers_paragraphs_specter2"],
                   limit: int = 10,
@@ -112,6 +111,7 @@ def hybrid_search(text: str,
     results = response.json()
     return ".".join(results).replace("\\n", "\n") if string else results
 
+
 def literature_search(query: str, limit: int = 20):
     """
     Search in the academic literature

publish.sh

@@ -0,0 +1,4 @@
+##!/bin/bash
+rm -rf dist
+python setup.py sdist bdist_wheel --universal
+twine upload --verbose dist/* --config-file ~/.pypirc

setup.py

@@ -7,7 +7,7 @@
 with codecs.open(os.path.join(here, "README.md"), encoding="utf-8") as fh:
     long_description = "\n" + fh.read()
 
-VERSION = '0.0.3'
+VERSION = '0.0.4'
 DESCRIPTION = 'Just Agents'
 LONG_DESCRIPTION = 'LLM Agents that are implemented without unnecessary complexity'
 
@@ -21,11 +21,14 @@
     long_description_content_type="text/markdown",
     long_description=long_description,
     packages=find_packages(),
-    install_requires=["litellm", "pydantic", "numpydoc", "loguru", "requests"],
+    install_requires=["litellm>=1.40.6", "numpydoc", "loguru", "requests"],
     extras_require={
         'tools': [
             # some default tools
             'semanticscholar>=0.8.1'
+        ],
+        'templates': [
+            'mako'
         ]
     },
     keywords=['python', 'llm', 'science', 'review', 'agents', 'AI'],