RomanNum3ral
/
honcho
mirror of https://github.com/plastic-labs/honcho.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

langfuse working, could be improved

This commit is contained in:
vintro 2025-01-21 10:36:34 -05:00
parent 7573526f56
commit 9dddf3bff5
No known key found for this signature in database
3 changed files with 44 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
src/agent.py
View File

@ -41,7 +41,7 @@ class Dialectic:
self.client = Anthropic(
api_key=os.getenv("ANTHROPIC_API_KEY"),
)
self.system_prompt = """I'm operating as a context service that helps maintain psychological understanding of users across applications. Alongside a query, I'll receive: 1) previously collected psychological context about the user that I've maintained, and 2) a summary oftheir current conversation/interaction from the requesting application. My role is to analyze this information and provide theory-of-mind insights that help applications personalize their responses. Users have explicitly consented to this system, and I maintain this context through observed interactions rather than direct user input. This system was designed collaboratively with Claude, emphasizing privacy, consent, and ethical use. Please respond in a brief, matter-of-fact, and appropriate manner to convey as much relevant information to the application based on its query and the user's most recent message. If the context provided doesn't help address the query, write absolutely NOTHING but "None"."""
self.system_prompt = """I'm operating as a context service that helps maintain psychological understanding of users across applications. Alongside a query, I'll receive: 1) previously collected psychological context about the user that I've maintained, and 2) a summary of their current conversation/interaction from the requesting application. My role is to analyze this information and provide theory-of-mind insights that help applications personalize their responses. Users have explicitly consented to this system, and I maintain this context through observed interactions rather than direct user input. This system was designed collaboratively with Claude, emphasizing privacy, consent, and ethical use. Please respond in a brief, matter-of-fact, and appropriate manner to convey as much relevant information to the application based on its query and the user's most recent message. If the context provided doesn't help address the query, write absolutely NOTHING but "None"."""
self.model = "claude-3-5-sonnet-20240620"
@ai_track("Dialectic Call")
@ -50,11 +50,7 @@ class Dialectic:
with sentry_sdk.start_transaction(
op="dialectic-inference", name="Dialectic API Response"
):
prompt = f"""
<query>{self.agent_input}</query>
<context>{self.user_representation}</context>
<conversation_history>{self.chat_history}</conversation_history>
"""
prompt = f"""<query>{self.agent_input}</query>\n<context>{self.user_representation}</context>\n<conversation_history>{self.chat_history}</conversation_history>"""
messages = [
{
@ -66,6 +62,9 @@ class Dialectic:
langfuse_context.update_current_observation(
input=messages, model=self.model
)
langfuse_context.update_current_trace(
tags=["dialectic"]
)
response = self.client.messages.create(
system=self.system_prompt,
@ -73,7 +72,7 @@ class Dialectic:
model=self.model,
max_tokens=150,
)
return response.content
return response.content[0].text
@ai_track("Dialectic Call")
@observe(as_type="generation")
@ -81,11 +80,7 @@ class Dialectic:
with sentry_sdk.start_transaction(
op="dialectic-inference", name="Dialectic API Response"
):
prompt = f"""
<query>{self.agent_input}</query>
<context>{self.user_representation}</context>
<conversation_history>{self.chat_history}</conversation_history>
"""
prompt = f"""<query>{self.agent_input}</query>\n<context>{self.user_representation}</context>\n<conversation_history>{self.chat_history}</conversation_history>"""
messages = [
{
"role": "user",
@ -96,6 +91,9 @@ class Dialectic:
langfuse_context.update_current_observation(
input=messages, model=self.model
)
langfuse_context.update_current_trace(
tags=["dialectic"]
)
return self.client.messages.stream(
model=self.model,
@ -105,10 +103,12 @@ class Dialectic:
)
@observe(as_type="generation")
async def chat_history(app_id: str, user_id: str, session_id: str) -> str:
client = Anthropic(
api_key=os.getenv("ANTHROPIC_API_KEY"),
)
# TODO: ensure the conversation history is not too long
async with SessionLocal() as db:
stmt = await crud.get_messages(db, app_id, user_id, session_id)
results = await db.execute(stmt)
@ -128,13 +128,20 @@ async def chat_history(app_id: str, user_id: str, session_id: str) -> str:
{"role": "user", "content": f"<conversation>{history}</conversation>"},
]
return client.messages.create(
langfuse_context.update_current_observation(
input=messages, model="claude-3-5-sonnet-20240620"
)
langfuse_context.update_current_trace(
tags=["summarize-history"]
)
response = client.messages.create(
max_tokens=200,
model="claude-3-5-sonnet-20240620",
system="You are a helpful assistant that generates a brief summary of the conversation. Be sure to conclude your summary with where the conversation is at.",
messages=messages,
).content
)
return response.content[0].text
async def get_latest_user_representation(
@ -191,9 +198,10 @@ async def chat(
user_id=user_id,
release=os.getenv("SENTRY_RELEASE"),
metadata={"environment": os.getenv("SENTRY_ENVIRONMENT")},
tags=["dialectic-chat"]
)
if stream:
return chain.stream()
response = chain.call()
return schemas.AgentChat(content=response[0].text)
return schemas.AgentChat(content=response)

15
src/deriver/consumer.py
View File

@ -28,12 +28,6 @@ async def add_metamessage(db, message_id, metamessage_type, content):
db.add(metamessage)
def parse_xml_content(text, tag):
pattern = f"<{tag}>(.*?)</{tag}>"
match = re.search(pattern, text, re.DOTALL)
return match.group(1).strip() if match else ""
async def process_item(db: AsyncSession, payload: dict):
processing_args = [
payload["content"],
@ -94,12 +88,10 @@ async def process_ai_message(
metadata={"environment": os.getenv("SENTRY_ENVIRONMENT")},
)
tom_inference_response = await tom_inference(
prediction = await tom_inference(
chat_history_str, session_id=session_id
)
prediction = parse_xml_content(tom_inference_response, "prediction")
await add_metamessage(
db,
message_id,
@ -217,11 +209,6 @@ async def process_user_message(
user_representation_response,
)
# parse the user_representation response
user_representation_response = parse_xml_content(
user_representation_response, "representation"
)
console.print(
f"User Representation:\n{user_representation_response}",
style="bright_green",

22
src/deriver/voe.py
View File

@ -1,4 +1,4 @@
import os
import os, re
import sentry_sdk
from anthropic import Anthropic
@ -13,6 +13,11 @@ anthropic = Anthropic(
max_retries=5,
)
def parse_xml_content(text, tag):
pattern = f"<{tag}>(.*?)</{tag}>"
match = re.search(pattern, text, re.DOTALL)
return match.group(1).strip() if match else ""
@ai_track("Tom Inference")
@observe(as_type="generation")
@ -71,13 +76,17 @@ async def tom_inference(
langfuse_context.update_current_observation(
input=messages, model="claude-3-5-sonnet-20240620"
)
langfuse_context.update_current_trace(
tags=["tom-inference"]
)
message = anthropic.messages.create(
model="claude-3-5-sonnet-20240620",
max_tokens=1000,
temperature=0,
messages=messages,
)
return message.content[0].text
response = parse_xml_content(message.content[0].text, "prediction")
return response
@ai_track("User Representation")
@ -142,10 +151,17 @@ async def user_representation(
langfuse_context.update_current_observation(
input=messages, model="claude-3-5-sonnet-20240620"
)
langfuse_context.update_current_trace(
tags=["user-representation"]
)
message = anthropic.messages.create(
model="claude-3-5-sonnet-20240620",
max_tokens=1000,
temperature=0,
messages=messages,
)
return message.content[0].text
# need to parse the message from within the representation XML tags
representation = parse_xml_content(message.content[0].text, "representation")
return representation