MindMapper

MindMapper

Python simulation where LLM-backed agents interact, maintain thought trees, and track sentiment over time. Each agent has a goal, a tree of thoughts, and uses keyword extraction (YAKE) and sentiment analysis (TextBlob) to update its internal state. Agents generate responses via the Anthropic API and look up entities through the Google Knowledge Graph API.

Status

Proof of concept. Runs but requires manual API key insertion in source code. No tests, no CLI arguments, no config file.

What it does

1. Creates N agents, each with a goal string and a thought tree (rooted tree of ThoughtNode objects).
2. On each simulation tick, agents interact with each other:
   • Generate a response using Anthropic Claude.
   • Extract keywords from the response (YAKE).
   • Analyze sentiment (TextBlob).
   • Update emotional state, interpersonal state, and knowledge graph.
   • Expand the thought tree with new child nodes.
3. The thought tree can be explored via depth-first, breadth-first, or heuristic search.
4. Simulation runs for a configurable number of iterations.

Components

File	Purpose
main.py	Entry point, creates agents and runs the simulation loop
agents.py	Agent class with LLM calls, sentiment, keyword extraction, thought tree
simulation.py	Simulation orchestrator
thought_tree_explorer.py	DFS, BFS, and heuristic traversal of thought trees
theory_of_mind.py	Stub for theory-of-mind reasoning between agents
decision_model.py	Decision logic
evaluation_metric.py	Metrics for evaluating agent understanding
interactive_interface.py	Console-based interaction (if used)
nlp_messaging.py	NLP utilities for agent messaging
task_scenario.py	Task/scenario definitions

Requirements

Dependency	Used for
anthropic	LLM calls (Claude)
textblob	Sentiment analysis
yake	Keyword extraction
requests	Google Knowledge Graph API
goo