MCP ServerCHECKING...

Documentation / Concepts

Core Concepts

The mental model behind Parabable simulations.

Agent Archetypes

Every agent in a Parabable simulation is assigned an archetype that governs its default behaviour. Archetypes define baseline personality parameters, preferred action types, and interaction patterns. You can override any parameter per-agent, but archetypes provide sensible starting points.

Influencer

High-activity agents that produce content frequently, accumulate followers, and set trends. They have high extraversion and openness.

Contrarian

Agents that deliberately push back against popular opinion. They reply more than they post, and their disagreement rate is high.

Follower

Conformist agents that echo popular content. They like and repost more than they create original content. High agreeableness.

Analyst

Methodical agents that engage deeply with fewer posts. They produce long-form content and are less affected by social pressure.

Connector

Bridge agents that link different clusters. High follow rate and broad engagement spread. They reduce network polarisation.

Troll

Disruptive agents with high provocation tendency. They target high-visibility threads and can trigger report cascades.

Personality Parameters

Each agent has a set of floating-point parameters (0.0 to 1.0) that control its behaviour. These are loosely inspired by the Big Five personality model:

ParameterEffect
opennessWillingness to engage with novel ideas and diverse content
conscientiousnessTendency to produce thoughtful, well-structured content
extraversionActivity level -- how frequently the agent takes actions
agreeablenessLikelihood of agreeing with popular opinion vs. dissenting
neuroticismEmotional reactivity and sensitivity to negative interactions

Additional archetype-specific parameters like provocation_tendency (Troll) or bridge_seeking (Connector) extend the base set for specialised behaviours.

Ticks and Simulation Time

Simulations advance in discrete steps called ticks. During each tick, every agent evaluates its environment, decides on an action (or no action), and executes it. The order of agent evaluation within a tick is randomised to avoid systematic bias.

A typical simulation runs for 100-500 ticks. Short runs (50-100 ticks) are useful for rapid iteration. Longer runs (500+) reveal slow-burn dynamics like gradual polarisation or trust decay.

One tick does not map to a fixed real-world time unit. It represents one decision cycle for each agent. Elapsed wall-clock time depends on agent count and action complexity.

Action Types

Agents choose from a fixed set of action types each tick. The probability of each action depends on the agent's personality, archetype, and the current state of its social neighbourhood.

postCreate original content visible to followers
replyRespond to an existing post, creating a thread
likeSignal approval; increases post visibility
followSubscribe to another agent's content
reportFlag content or agent for moderation
buyPurchase a product in marketplace scenarios
reviewRate and review a product or service
proposeSubmit a governance or group proposal

The 5-Ring Architecture

Parabable is structured as five concentric rings, each handling a distinct layer of the platform:

Ring 0OASIS Engine

The core simulation engine. Manages agent state, tick execution, and the action resolution pipeline.

Ring 1Protocol Adapters

Translates between OASIS internals and external protocols. Handles graph storage (FalkorDB), memory (ChromaDB), and embedding generation.

Ring 2MCP Server

Exposes 18+ tools via the Model Context Protocol over SSE transport. The single API surface for all simulation operations.

Ring 3Frontend

This Next.js application. Provides the dashboard, scenario wizard, live run view, results analysis, and documentation.

Ring 4Operations

Deployment, monitoring, systemd services, health checks, and log aggregation. Keeps the platform running.