Axiom Engine

Axiom Engine is a modular autonomous research system combining:

• Dreamer-style RSSM world model
• latent actor-critic agents
• PettingZoo-style multi-agent coordination
• symbolic communication and soft rule reasoning
• knowledge graph based scientific loop
• Ray-style queue and worker infrastructure
• meta-learning and bounded self-improvement primitives

Directory Structure

core/
  world_model.py        RSSM, Dreamer-style wrapper, imagination rollout
  agent.py              latent actor, critic, centralized critic
  multi_agent.py        multi-agent shared-environment coordinator
  language.py           symbolic communication channel
  reasoning.py          soft rule engine

science/
  knowledge_graph.py    concept nodes and relation edges
  hypothesis.py         hypothesis generation
  experiment.py         experiment planning and simulation execution
  analyzer.py           result analysis and acceptance

infrastructure/
  orchestrator.py       autonomous research loop
  worker.py             long-running task worker
  queue.py              bounded task queue
  storage.py            JSONL persistence

optimization/
  meta_learning.py      genome and evolution strategy
  rsi.py                proposal-validation self-improvement loop

utils/
  config.py             runtime config
  metrics.py            metric accumulator
  logging.py            logging setup

src/
  Existing Blender scene compiler vertical slice.

Install

python3 -m venv .venv
.venv/bin/python -m pip install -r requirements.txt

Run Axiom Engine

.venv/bin/python main.py --iterations 3

Output includes:

• iterations executed
• accepted/rejected experiment count
• resulting knowledge graph
• experiment records

Run Tests

.venv/bin/python -m pytest -q

Current expected result:

15 passed

Current Status

Implemented:

• RSSM world model with prior/posterior latent dynamics
• Dreamer-style imagination rollout interface
• actor-critic agent operating in latent space
• multi-agent coordinator using a PettingZoo-style protocol
• symbolic speaker/listener module
• soft rule engine
• knowledge graph add/query
• hypothesis -> experiment -> analyze -> graph update loop
• bounded task queue and worker abstraction
• JSONL persistence
• meta-learning genome and RSI validation loop
• Blender compiler vertical slice remains intact under src/

Reference repos are cloned outside this repository at:

/Users/fallofpheonix/Project/axiom-reference-repos

They are reference components, not vendored source.