ActorScope-AI: A Multi-Agent Stakeholder Analysis and Decision Simulation Framework Based on Large Language Models

ActorScope-AI is a multi-agent simulation framework designed for analyzing complex stakeholder scenarios. It leverages LangGraph for orchestration, Ollama for local LLM reasoning, and Mem0 for persistent memory to enable structured modeling and transparent reasoning of organizational dynamics. Its core goal is to address key questions about stakeholder goals, potential actions, critical tensions, reasonable outcomes, and effective intervention strategies in multi-party scenarios.

Traditional analysis methods often simplify complex multi-stakeholder dynamics into static models or single-perspective summaries, failing to capture subtle changes and emergent behaviors. ActorScope-AI targets this gap by providing a dedicated multi-agent simulation platform that explicitly models participants, relationships, environment, and scenario states to deepen understanding of complex organizational dynamics.

ActorScope-AI uses a 3-layer state separation model:

1. Long-running state (actors, relationships, environment, event history, evaluation)
2. Per-round temporary state (interpretations, salience scores, selected actors, action proposals)
3. Execution state (run ID, current round, stop conditions)

Key technologies include LangGraph (orchestration), Ollama (local LLM backend), Mem0 (persistent memory), and Pydantic (typed state modeling). It balances LLM-driven reasoning (interpretation/action selection) with deterministic state updates for precise control.

1. Multi-dimensional participant modeling: Each actor has roles, base objectives, priorities, constraints, capabilities, red lines, and interaction styles.
2. Directed relationship modeling: Encodes asymmetric dynamics like trust, alignment, conflict, dependency, and influence.
3. Mixed round/turn loop: All actors interpret the scene first, then a primary actor is selected to act (balances comprehensiveness and efficiency).

ActorScope-AI applies to:

• Organizational strategy analysis (simulate stakeholder reactions to strategic moves)
• Negotiation preparation (model opponent teams and test concession strategies)
• Geopolitical/policy analysis (track alliance evolution and conflict points)
• Project management (identify change resistance and optimize communication strategies)

• Observability: Generates structured outputs (trace.jsonl, final_output.json, summary.md) for decision tracing.
• Transparency: Uses decision-level Architecture Decision Records (ADRs) to document design choices.
• Persistent memory: Integrates Mem0 for actor/relationship/scenario pattern memory across runs, with distillation and retrieval to manage long-term insights.

• Completed: Typed state model, LangGraph execution loop, LLM-driven interpretation/action selection, deterministic updates, decision tracking, ADRs.
• To-do: Enhance memory utilization, expand scenario coverage, add UI layer, improve evaluation logic.
• Principles: Domain-agnostic core, state-driven execution, single-writer mutable state, deterministic world mutation, observability as first-class concern.

ActorScope-AI offers a new paradigm for applying LLMs to organizational analysis. It helps decision-makers (strategists, policy analysts, negotiators) pre-simulate multi-party interactions, identify tensions/opportunities, and test interventions. While not replacing human judgment, it enhances situational understanding and forward-looking capabilities, with potential to become a key tool in organizational intelligence.