lore / lore_generator.py
StillWorking's picture
Upload 16 files
a387ea0 verified
Raw History Blame Contribute Delete
3.79 kB
"""
Lore Generator – LangChain + Groq (llama3-70b-8192) with Gemini fallback
Accepts a list of 2–6 NodeRef objects and weaves them into a single lore narrative.
Orchestration order:
1. cascade_config β†’ verify budget & latency constraints
2. memory_manager β†’ recall any prior lore for this entity group
3. Groq Llama 3 70B β†’ generate new lore (ultra-fast via Groq API)
4. memory_manager β†’ retain the generated lore
"""
from __future__ import annotations
import os
from typing import TYPE_CHECKING
from langchain_groq import ChatGroq
from langchain_google_genai import ChatGoogleGenerativeAI # Kept for future fallback!
from langchain_core.prompts import PromptTemplate
from fastapi import HTTPException
from agents.cascade_config import check_constraints, record_call
from agents.memory_manager import recall, retain
if TYPE_CHECKING:
# NodeRef is a Pydantic model defined in app.py; imported only for type hints
# to avoid a circular import at runtime.
from app import NodeRef
async def generate_lore(nodes: list) -> str:
"""Generate lore that weaves together all supplied nodes (2–6 entities).
Args:
nodes: list of NodeRef-like objects with `.name` and `.type` attributes.
"""
# 1. Load keys (Keeping both in the codebase as requested)
groq_key = os.getenv("GROQ_API_KEY")
google_key = os.getenv("GOOGLE_API_KEY") # Kept for fallback
if not groq_key:
raise HTTPException(status_code=500, detail="Groq API Key missing in .env file.")
# 2. Governance check
check_constraints()
# 3. Build a stable memory key from all node names (sorted for consistency)
sorted_names = sorted(n.name.lower() for n in nodes)
memory_key = "::".join(sorted_names)
prior = recall(memory_key)
memory_context = prior if prior else "No prior history recorded."
# 4. Dynamically format the entity list for the prompt
entities_text = "\n".join(
f"{i}. {node.name} (Type: {node.type})"
for i, node in enumerate(nodes, 1)
)
try:
print("\n--- GENERATING LORE WITH GROQ (Llama 3) ---")
# 5. Connect to Groq instead of Gemini
llm = ChatGroq(
model="openai/gpt-oss-120b", # Using the newer, more powerful 120B model
api_key=groq_key,
temperature=0.8
)
template = (
"You are a master fantasy worldbuilder. The user has selected a web of "
"interconnected entities from their world.\n\n"
"Entities to connect:\n{entity_list}\n\n"
"Prior context: {memory_context}\n\n"
"Write a cohesive, engaging, 3-paragraph piece of lore that organically "
"weaves all of these specific characters, locations, events, and organizations "
"together into a single narrative or historical record. Make it mysterious and "
"highly detailed. Ensure every entity listed above plays a meaningful role in "
"the story. Do not use markdown formatting."
)
prompt = PromptTemplate.from_template(template)
chain = prompt | llm
response = await chain.ainvoke({
"entity_list": entities_text,
"memory_context": memory_context,
})
lore = response.content
# 6. Persist to memory
retain(memory_key, lore)
# 7. Record usage for cascadeflow budget tracking
record_call()
print("Success! Lore generated instantly.")
print("-------------------------------------------\n")
return lore
except HTTPException:
raise
except Exception as e:
print(f"Groq API Error: {str(e)}")
raise HTTPException(status_code=500, detail="Failed to connect to Groq API.")