mgia 0.2.1

MGIA 4.0 — Strategic Macro-Geopolitical Intelligence OS (autonomous agent)

MGIA 4.0 — Strategic Macro-Geopolitical Intelligence OS

Autonomous intelligence agent built on the principle:

The LLM is not the system's intelligence. It is an analytical interface over an intelligence infrastructure. DATA → EVIDENCE → CAUSALITY → SCENARIOS → ADVERSARIAL → MEMORY → ANALYSIS

See ARCHITECTURE.md for the full technical reference and magi v4.txt for the source spec.

Status

v0.2.1 — platform complete, first PyPI release. All 6 layers + orchestration, governance, observability, live data, and delivery surfaces. 233 tests passing (ruff + mypy clean).

Layer / Component	Status
L1 Data Collection (JSON file · RSS/Atom · JSON-API/NewsAPI · Reddit · GDELT)	✅
L2 Trust & Validation	✅
L3 Causal Intelligence	✅
L4 Scenario Intelligence	✅
L5 Institutional Memory	✅
L6 Strategic Analysis (Mock + real DeepSeek/OpenAI-compat LLM)	✅
Observability + Governance engines	✅
Master Orchestrator + CLI	✅
Reproducibility cache · fault-tolerant collection	✅
HTTP serve (auth · rate-limit · HTML dashboard)	✅
Report export (Markdown · HTML · PDF)	✅
Packaged wheel/sdist (BUSL-1.1)	✅

Verified on a live provider

In addition to the mock-based automated test suite, MGIA has been validated end-to-end against a live DeepSeek endpoint.

The validation exercised the complete six-layer pipeline, including governance (validated=True). Numerical metrics (including confidence) are produced deterministically by the pipeline, while the LLM is used only to generate the narrative report.

See docs/live-smoke-v0.2.1.md.

Quickstart

cd projects/magi
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"          # or: make install
python -m pytest -q               # 233 tests

CLI Usage

# Ask a strategic question (full 6-layer pipeline)
python -m mgia ask "What are the implications of BRICS expansion?"

# Deployment mode / output formats
python -m mgia ask "Question?" --mode institutional
python -m mgia ask "Question?" -f json
python -m mgia ask "Question?" -f html -o report.html
python -m mgia ask "Question?" -f pdf  -o report.pdf

# Load a config file (live feeds, cache, source APIs, LLM provider)
python -m mgia --config examples/config.json ask "Question?"

# Reproducibility check (same input twice)
python -m mgia --config examples/config.json verify "Question?"

# System status / institutional memory / governance lint
python -m mgia status
python -m mgia memory
python -m mgia lint

# HTTP API (optional auth + rate limit); dashboard at /dashboard
python -m mgia --config examples/config.json serve --port 8080 \
    --token "$MGIA_API_TOKEN" --rate-limit 60

# Version
python -m mgia --version

Real LLM (DeepSeek / OpenAI-compatible): set MGIA_LLM_API_KEY (or DEEPSEEK_API_KEY) and use --config examples/config.deepseek.json.

Python API

from mgia.config import MGIAConfig, DeploymentMode
from mgia.orchestrator import MGIA

# Initialise with defaults
mgia = MGIA()

# Full pipeline
report, markdown = mgia.ask("What are the implications of BRICS expansion?")
print(report.confidence_summary.overall_confidence)
print(markdown[:200])

# System status
status = mgia.status()
print(status["layers"])

# Governance checks
lint = mgia.lint()
print(lint["all_passed"])

# Memory summary
mem = mgia.memory_status()
print(mem["learning_count"], "learnings stored")

Deployment Modes

Mode	Red Team	Scenario Killers	Governance
calibration	50%	aggressive	relaxed
standard	35%	normal	relaxed
institutional	100%	mandatory	strict

Configuration

Via MGIAConfig (pydantic model, load from JSON, env vars, or defaults):

from mgia.config import MGIAConfig, DeploymentMode

config = MGIAConfig(
    mode=DeploymentMode.STANDARD,
    data_dir="examples",
    memory_path="memory/custom_memory.json",
    log_dir="reports/logs",
)

# Or from environment variables: MGIA_MODE, MGIA_DATA_DIR, etc.
config = MGIAConfig.from_env()

# Or from file:
config = MGIAConfig.from_file("config.json")

Project Structure

projects/magi/
  ARCHITECTURE.md            Technical reference
  IMPLEMENTATION_PLAN.md     Step-by-step build plan
  magi v4.txt                Source specification (frozen)
  README.md                  This file
  pyproject.toml             Package metadata + CLI entry point
  LICENSE                    Business Source License 1.1
  CHANGELOG.md               Release notes
  Makefile                   test / lint / typecheck / run / serve / build
  mgia/
    __init__.py              __version__
    config.py                MGIAConfig, DeploymentMode, LLMConfig
    orchestrator.py          MGIA — master orchestrator
    cli.py                   CLI: ask, status, memory, lint, verify, serve
    server.py                HTTP API (serve) + dashboard + auth/rate-limit
    report_export.py         Markdown/HTML/PDF export
    observability.py         ObservabilityEngine, MetricsCollector
    governance.py            GovernanceEngine, GovernancePolicy
    layer1_data/             Sources (file · RSS · JSON-API/NewsAPI · Reddit · GDELT · cache), collector
    layer2_trust/            Source trust, data quality, bias auditor
    layer3_causal/           Event graph engine, confidence, builder
    layer4_scenario/         Bayesian, adversarial, scenario killer, orchestrator
    layer5_memory/           Pattern store, learning loop, persistence
    layer6_strategic/        LLM interface (Mock + OpenAICompatBrain), report builder
  examples/                  config.json, config.deepseek.json, config.live.json, …
  tests/                     test suite

License

MGIA 4.0 is licensed under the Business Source License 1.1 (BUSL-1.1) — see LICENSE.

• Source-available: you may copy, modify, and make non-production use freely.
• Production use requires a commercial license from the Licensor (ooopalladiumsb) until the Change Date — Additional Use Grant: None.
• On the Change Date (2030-07-08), each released version automatically converts to the Apache License 2.0.

For commercial-license inquiries, contact the Licensor.