Class: Llmemory::Extractors::FactExtractor

Inherits:
Object
  • Object
show all
Defined in:
lib/llmemory/extractors/fact_extractor.rb

Instance Method Summary collapse

Constructor Details

#initialize(llm: nil) ⇒ FactExtractor

Returns a new instance of FactExtractor.



8
9
10
 
# File 'lib/llmemory/extractors/fact_extractor.rb', line 8

def initialize(llm: nil)
  @llm = llm || Llmemory::LLM.client
end

Instance Method Details

#classify_item(content) ⇒ Object 



45
46
47
48
49
50
51
52
53
54
 
# File 'lib/llmemory/extractors/fact_extractor.rb', line 45

def classify_item(content)
  return "general" if content.to_s.strip.empty?
  prompt = <<~PROMPT
    Classify this fact into ONE category. Use lowercase with underscores. Examples: work_life, personal_life, preferences, general.
    Fact: #{content}
    Return ONLY the category name, nothing else.
  PROMPT
  result = @llm.invoke(prompt.strip).to_s.strip.downcase.gsub(/\s+/, "_")
  result.empty? ? "general" : result
end

#evolve_summary(existing:, new_memories:) ⇒ Object 



25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
 
# File 'lib/llmemory/extractors/fact_extractor.rb', line 25

def evolve_summary(existing:, new_memories:)
  memory_list_text = Array(new_memories).map { |m| "- #{m}" }.join("\n")
  prompt = <<~PROMPT
    You are a Memory Synchronization Specialist.
    Topic Scope: User Profile

    ## Original Profile
    #{existing.to_s.empty? ? "No existing profile." : existing}

    ## New Memory Items to Integrate
    #{memory_list_text}

    # Task
    1. Update: If new items conflict with the Original Profile, overwrite the old facts.
    2. Add: If items are new, append them logically.
    3. Output: Return ONLY the updated markdown profile.
  PROMPT
  @llm.invoke(prompt.strip).to_s
end

#extract_items(conversation_text) ⇒ Object 



12
13
14
15
16
17
18
19
20
21
22
23
 
# File 'lib/llmemory/extractors/fact_extractor.rb', line 12

def extract_items(conversation_text)
  prompt = <<~PROMPT
    Extract discrete facts from this conversation.
    Focus on preferences, behaviors, and important details.
    Conversation: #{conversation_text}
    Return as JSON array of objects with "content" and "importance" (0-1) keys.
    Importance: 0.8-0.95 for preferences/corrections/decisions, 0.5-0.8 for factual context, 0.3-0.5 for ephemeral.
    Example: [{"content": "User prefers Ruby", "importance": 0.9}, {"content": "User mentioned the weather", "importance": 0.4}]
  PROMPT
  response = @llm.invoke(prompt.strip)
  parse_items_response(response)
end