aGLM — Autonomous General Learning Model

A logic-augmentation layer for language models — not a model itself.

aGLM wraps any existing model with Socratic, epistemic, nonmonotonic, and BDI reasoning. Because the reasoning layer is pure Python, the interfaces load instantly and the language model is optional and pluggable.

---

Overview

AGLM is a hybridization of MASTERMIND aGLM with RAGE, distilled from the Professor Codephreak / easyAGI research lineage. Its premise is simple:

AGLM is not a language model — it augments one. Every input can be run through a cognitive pipeline (Socratic framing → epistemic belief → nonmonotonic defeasibility → BDI plan) that produces a logic-augmented prompt. That prompt is what gets handed to a model, or, with no model attached, is the deliverable itself.

The repository contains three things:

1. Two consoles — working user interfaces that actually load.
2. A modern aglm/ Python package — the installable Perceive-Decide-Act core.
3. The historical research modules — the cognitive faculties (socratic.py, bdi.py, logic.py, MASTERMIND.py, …) that AGLM grew out of.

---

Quickstart — the consoles

Both front ends run with zero heavy ML dependencies. The language model runs in Ollama (local for modest hardware, :cloud for larger reasoning) and is entirely optional.

# 1) Flask cognitive console — every faculty, zero heavy deps
./run_console.sh              # → http://localhost:5000

# 2) AI SDK participant console — streaming chat on the Vercel AI SDK v7
./run_aisdk_console.sh        # Flask brain :5000 + Next.js console :3000

The AI SDK console adds a streaming chat with a model picker (local + :cloud), an Advanced model-state panel (temperature, top-p, top-k, penalties, context window, seed), a .history tab, and a live token counter.

Faculty	Module	In the console
Socratic	reasoning.py	premises · challenge · conclude · probing questions
Logic tables	logic.py	truth tables over boolean expressions
Nonmonotonic	nonmonotonic.py	default logic with retractable conclusions
Epistemic	epistemic.py	belief state + revision
BDI	bdi.py	Belief / Desire / Intention agent state
Autonomize	autonomize.py	self-healing retry loop
MASTERMIND	MASTERMIND.py	agent load · validate · execute
Memory	memory.py	persist + browse ./memory/*.json
Prediction	prediction.py	optional sklearn-style inference

📖 CONSOLE.md — usage walkthrough  ·  🧭 TECHNICAL.md — architecture  ·  🛠 .claude/skills/aisdk/ — AI SDK v7 reference

---

The modern aglm/ package

A standalone, Apache-2.0 Python distribution of aGLM's autonomous-learning loop, distilled from the agenticplace/mindX pattern.

pip install .                  # core only
pip install ".[rage]"          # with GATERAGE/RAGE
pip install ".[mastermind]"    # with GATERAGE/mastermind
pip install ".[dev]"           # pytest, ruff

import asyncio
from aglm import AGLMCore, AutonomousLoop, Decision, PerceptionContext

async def perceive():          return PerceptionContext(facts={"hour": 14})
async def decide(ctx, beliefs): return Decision(action="log", args=ctx.facts)
async def act(decision):        return {"success": True}

async def main():
    core = AGLMCore(perceive=perceive, decide=decide, act=act)
    print(await core.cycle())                              # one cycle
    loop = AutonomousLoop(core, interval_seconds=300.0)    # or a periodic runner
    await loop.start()
    await loop.stop()

asyncio.run(main())

Module	Class	Responsibility
aglm/core.py	AGLMCore	Perceive · Orient · Decide · Act cycle
aglm/beliefs.py	BeliefSystem	claim + confidence + source attribution
aglm/cycle.py	AutonomousLoop	periodic runner with circuit breaker

pip install ".[dev]" && pytest -v

Canonical contract: docs/aglm_as_a_service.md.

---

Ecosystem

RAGE remembers · aGLM decides · MASTERMIND orchestrates.

• GATERAGE/RAGE — retrieval substrate (memory)
• GATERAGE/mastermind — strategic orchestrator (planning)
• pythaiml/automindx — Professor Codephreak deployment environment

---

Historical research modules

The files at the repo root document the philosophical foundation aGLM grew out of and remain preserved unchanged. The cognitive faculties are used directly by the consoles above.

Module	Purpose
MASTERMIND.py	Core controller — loads, validates, and executes agents concurrently
bdi.py	Beliefs · Desires · Intentions agent framework
socratic.py / reasoning.py	Socratic question-and-answer reasoning over premises
logic.py	Formal logic operations and truth tables
nonmonotonic.py	Non-monotonic reasoning — beliefs adapt to contradicting evidence
epistemic.py	Knowledge/belief tracking and revision over time
autonomize.py	Self-healing autonomy with exponential backoff
prediction.py	Forecasting via statistical / ML models
terminai.py · terminai_module.py	OpenAI command-mode (cmd:) integration
SimpleCoder.py	Reusable code snippets and templates
config.json	Default allowed agency for MASTERMIND
chunk4096.py	Guards inputs above the 4096-token context of the original model

Note. The legacy Gradio entrypoints (uiux*.py, main.py) tried to load a multi-GB model before rendering and are kept for reference only. Use the consoles above instead — they are the working interfaces.

---

Attribution & license

• MASTERMIND — agent creator and control agent · © codephreak, GPLv3, 2024 (NFT).
• Modern aglm/ package — Apache-2.0, © 2024–2026 GATERAGE / Professor Codephreak.
• Project details: rage.pythai.net.

See LICENSE for the full text.