Skip to content

DDA-X

A framework and experimental lab for stateful, identity-persistent LLM agents.

DDA-X treats surprise as something that can contract an agent before it expands. In this repository, that idea is implemented across simulation work, memory systems, LLM control logic, and a live Discord agent scaffold.

This docs site covers six connected surfaces:

  • the framework itself
  • the simulation lineage that developed it
  • the applied systems surface for companion builds and interfaces
  • the applied Discord agent runtime
  • the support utilities used to inspect, verify, and visualize runs
  • the archive and taxonomy surfaces that explain the repo's structure and history

Core Idea

Standard agent framing often treats surprise as fuel for exploration.

DDA-X explores a different path:

surprise -> rigidity -> contraction

That contraction can then affect:

  • state update magnitude
  • response bandwidth
  • memory salience
  • identity drift resistance
  • reply selection behavior

What This Repo Contains

Framework

The mathematical and architectural framing lives in:

Simulation Lineage

The simulation corpus tracks the growth of the ideas across many experiments, from simple mechanics demos to richer multi-agent and live-system scenarios.

Applied Systems

Some DDA-X builds are better understood as usable software surfaces than as raw simulations.

Discord Agent

The Discord runtime takes the framework into a persistent live agent:

  • stateful session routing
  • memory retrieval
  • adaptive notes
  • telemetry export
  • profile-based identity anchoring

Setup and runtime notes live here:

Support Utilities

The repo also contains a dedicated support-tooling surface for provider probes, dashboards, verification scripts, and session analysis:

Archive and Taxonomy

The repo documents both its current structure and its historical layer:

Key Mechanisms

Mechanism Description
Multi-timescale rigidity Startle, stress, and trauma-like components can evolve differently
Surprise-weighted memory Retrieval combines similarity, recency, and salience
Identity persistence Agent state is pulled toward a stable identity attractor
Wound activation Semantic and lexical triggers can amplify contraction
Rigidity-bound output Internal state affects the style and bandwidth of responses

Recommended Entry Points

If you want theory:

If you want implementation:

If you want the lineage:

If you want the structural map:

If you want support utilities:

If you want the archive layer:

If you want the live application:

If you want the applied systems surface:

Quick Start

python simulations/simulate_agi_debate.py

python apps/brobot/true_brobot.py

python run_discord_agent.py

Notes

  • The chronology pages document a curated, analyzed lineage of DDA-X work.
  • The simulations/ directory also contains newer branches, support scripts, and side experiments beyond any single frozen count.
  • Stronger user-facing builds can move into apps/ when they deserve first-class software status.
  • scripts/ is the active support surface; archive/ is the preserved historical surface.
  • The refined master prompt is a synthesis artifact from the project's development, not the only source of truth for the repo.