LLM AI integration

[dgtlmoon]

Closes #3204

• How to add LLM prompts but not feel overwhelmed by the interface (too many options, buttons, etc etc).. feels messy still
• What happens when someone wants to use Gemini instead of OpenAI? (maybe 1 global setting instead?)
• Does 'groups' work with the prompts?
• how would it work with Browser Steps?
• maybe add a textarea to "Filters" so you could add some text like "Show me just the text inside of the recipe"
• Could work with notification body
• in Filters/Text, does the LLM run BEFORE or AFTER the filters?

the rest of the work really is just a requests request..

in the edit..

Settings

---

langchain memory could be cool too, compare and summarise from the previous check

from langchain.chat_models import ChatOpenAI
from langchain.schema import AIMessage

# New scraped text
new_text = "This is the updated web page content with new info."

# Retrieve old text from memory
previous_msgs = memory.chat_memory.messages
previous_text = None

# Find the last page text stored
for msg in reversed(previous_msgs):
    if isinstance(msg, AIMessage):
        previous_text = msg.content
        break

if previous_text is None:
    previous_text = ""

# Create the comparison prompt
comparison_prompt = f"""
Compare the following web page versions and tell me:
- Did anything change?
- Summarize the differences.

=== OLD VERSION ===
{previous_text}

=== NEW VERSION ===
{new_text}
"""

llm = ChatOpenAI(model="gpt-4")

result = llm.predict(comparison_prompt)

print(result)