Simulation Manager Core features: 1. Read entities from graph and filter 2. Generate Wonderwall Agent Profiles 3. Use LLM to int

Minimal stand-in for SimulationState used by select_public_cards.

Tiny stand-in for SimulationState — only the fields the sitemap reads.

r"""Manual predefined actions that need to be executed by certain agents. Args: action_type: The action to perform. Can be an ``ActionTyp

Thin JSON-API client for a SearXNG instance.

Context manager that activates a locale for its body. Example:: with use_locale("zh-CN"): run_simulation(...)

r"""AgentGraph class to manage the social graph of agents.

Entity reading and filtering service (via GraphStorage / Neo4j) Main capabilities: 1. Read all nodes from the graph 2. Filter out me

r"""Agent that participates in a Wonderwall simulation. Supports both the legacy social-media workflow (``SocialAction`` + ``SocialEnvironmen

r"""Represents actions generated by a Language Learning Model (LLM).

Platform simulation result container

Task Manager Thread-safe task status management

Lightweight stand-in for SimulationState used by select_public_cards.

Manages belief tracking for a single platform's simulation.

Raised when embedding generation fails.

Graph Retrieval Tools Service (via GraphStorage / Neo4j) [Core Retrieval Tools - Optimized] 1. insight_forge - Deep Insight Retrieval (M

Raised when a SearXNG search request fails.

Simulation configuration intelligent generator Uses LLM to analyze simulation requirements, document content, and graph entity information

Minimal Flask-shaped request stub: ``.args`` + ``.headers``. Both are plain ``dict``-like with a ``.get(key, default=None)`` API, which match

LLM client using OpenAI-compatible APIs

Neo4j CE implementation of the GraphStorage interface.

Node Information

Report section

Wonderwall Agent Profile data structure

Wonderwall Profile Generator Convert entities from the knowledge graph to Agent Profile required by Wonderwall simulation Optimization

Per-process, per-channel ``(sim_id, status)`` dedup set. The runner's two terminal code paths both notify, so each channel must fire at most

Non-blocking JSONL event logger with SSE subscriber support.

Simulation IPC Client (used by the Flask side) Used to send commands to the simulation process and wait for responses

Agent action record

Single Agent Interview Result

Event configuration

Builds a knowledge graph through the GraphStorage interface.

Single-platform action logger

Report Agent - Simulation Report Generation Agent Uses ReACT (Reasoning + Acting) pattern: 1. Planning phase: Analyze simulation require

Report outline

Declarative definition of a simulation type. Bundle together the platform, action, environment, and prompt classes that define a simulation.

Simulation state

Accumulates round snapshots for the entire simulation.

Drop-in replacement for LLMClient that shells out to `claude -p`. Implements the same chat() / chat_json() interface.

r"""Clock used for the sandbox.

Edge Information

Tail a JSONL file from last-read position. Thread-safe.

Interview Result Contains interview responses from multiple simulated Agents

Shared state enabling feedback between prediction markets and social media. Usage: bridge = MarketMediaBridge() # In Polymarket

Ontology Generator Analyzes text content to generate entity and relationship type definitions

r"""Platform.

Platform-specific configuration. NOTE ON DEFAULT VALUES: The defaults below are heuristic starting points, NOT calibrated against real platfo

Complete report

Analyzes a simulation round and produces belief updates + feedback.

One round's data across all platforms.

Captured state after one simulation round.

Search Result

Simulation log manager Unified management of all log files, separated by platform

Simulation run state (real-time)

Result of an AMM trade.

Uses the LLM to research real-world context for entity personas. The LLM is treated as a knowledge base. If the configured model supports web

A dict whose ``.get`` raises for a configured key (precedence test).

Agent activity record

Activity configuration for a single Agent

Build, store, and update community clusters over an entity graph.

Detect and invalidate superseded edges at ingestion time.

Shared in-memory log of agent actions across all platforms. Safe for concurrent use within a single asyncio event loop (all callers shar

Generate embeddings using Ollama or OpenAI-compatible API.

Resolve newly-extracted entities against existing graph nodes.

Graph memory update service (via GraphStorage / Neo4j) Monitors simulation action logs and sends agent activities to the graph in real-time.

IPC command handler

IPC command handler

Deep Insight Retrieval Result (InsightForge) Contains retrieval results from multiple sub-questions and integrated analysis

Pool of MCP server processes usable during a simulation run. Construct once at simulation start, pass to agent observation code, :meth:`dispa

Parsed MCP server entry from the YAML manifest.

Current state of all prediction markets.

Filter out camel-ai warnings about max_tokens (we intentionally don't set max_tokens, letting the model decide)

Filter out camel-ai warnings about max_tokens (we intentionally don't set max_tokens, letting the model decide)

Filter out camel-ai warnings about max_tokens (we intentionally don't set max_tokens, letting the model decide)

Extract entities and relations from text using local LLM.

Breadth Search Result (Panorama) Contains all related information, including expired content

Dual-platform IPC command handler Manages environments for both platforms and handles Interview commands

Builds system prompts for prediction market trader agents.

Project Data Model

Project Status

Per-report buffer. Start a section, record steps, finalize (which flushes one section's rows to Neo4j atomically), move to the next section.

Builds the system prompt for a Reddit-style simulation.

Reddit simulation runner