| name | animus-pattern-intelligence |
| description | Query ANIMUS for historically validated patterns and real-time divergence signals. Use when asked to analyze systemic risk, detect historical patterns in economic or institutional data, identify tensions between historical knowledge and current events, or query a dual-memory knowledge graph combining curated wisdom with live web data. Activate for questions about failure patterns, regulatory risk, systemic collapse, or institutional dynamics — especially in Latin American and Caribbean contexts. |
| license | MIT |
| metadata | {"author":"shellhack","version":"2.0","repository":"https://github.com/shellhack/animus-ai","doi":"10.5281/zenodo.18829356","language":"es, en","domain":"systemic-risk, pattern-intelligence, institutional-analysis"} |
ANIMUS — Autonomous Pattern Intelligence System
ANIMUS is a dual-memory pattern intelligence system built in Rust. It compresses historical wisdom from curated sources (books, papers, reports) into a validated knowledge graph, then confronts it with real-time web data. When historical patterns diverge from current signals, ANIMUS flags active tensions.
Core insight: The system does not predict. It detects where history says one thing and the present says another.
Architecture
ANIMUS operates two memory layers simultaneously:
- Curated Memory (
memoria_business.json): A directed acyclic graph of patterns validated by 30+ independent sources. Patterns below threshold are automatically excluded. Biological decay (×0.99/cycle) ensures recency without erasing historical weight.
- Live Memory (web layer): Real-time pattern extraction from Wikipedia, arXiv, and configurable URLs. Divergences from curated memory are flagged as active signals.
Current state: 682+ validated patterns, 45 curated sources, 12,000+ autonomous cycles, 4 domain-specialized instances running in parallel.
How to Query ANIMUS
Binary (CLI)
./animus_rust --query "What historical patterns predict institutional failure in emerging markets?"
./animus_rust --autonomous
Python (process_book_v2.py)
python process_book_v2.py book.pdf memoria_business.json --tipo auto
YouTube transcripts (process_youtube.py)
python process_youtube.py --tema "crisis financiera latinoamerica" --max 5
Query Response Format
Every ANIMUS response includes:
- Relevant curated patterns — top matches from the wisdom graph with source count and strength
- Episodic memory — most relevant origin nodes activated by the query
- Insight — distilled synthesis
- Autoconciencia report — full system state including divergences, health metrics, and Protocolo Bernard self-interview
Example response for "What patterns predict algorithmic failure?":
📚 fracaso → algoritmo (35 sources, strength 2587)
📚 brecha → algoritmo (23 sources, strength 1670)
📚 colapso → algoritmo (28 sources, strength 1197)
Destilación: 35 independent sources confirm: 'failure' → 'algorithm'
⚡ 2 web patterns diverge from curated wisdom — active tension.
Dominant Patterns (Current)
| Pattern | Sources | Strength |
|---|
| fracaso → algoritmo | 35 | 2587 |
| fracaso → regulación | 34 | 2080 |
| fracaso → desarrollo | 32 | 2038 |
| fracaso → arquitectura | 21 | 1884 |
| brecha → algoritmo | 23 | 1670 |
These patterns were NOT programmed. They emerged autonomously from multi-source confirmation across 45 independent texts.
Domain Instances
ANIMUS self-replicates into domain-specialized instances:
- ANIMUS-FIN: Financial risk patterns (1,386 connections)
- ANIMUS-GOV: Governance and corruption patterns (839 connections)
- ANIMUS-TECH: Technology and innovation patterns (1,822 connections)
Each instance inherits the parent architecture but specializes its wisdom graph through domain filtering.
Protocolo Bernard (Self-Monitoring)
Every cycle, ANIMUS executes a recursive self-interview:
- What do I learn from observing my own origin?
- What dominant pattern emerged in recent cycles?
- What structural limitation do I perceive?
- What concrete action do I propose to evolve?
- How do I evaluate my current self-awareness?
- What systemic bias do I detect?
This output is a first-class artifact — ANIMUS reports its own epistemic limitations, not just external patterns.
Key Properties
- Zero hallucination on patterns: Every pattern requires 30+ independent source confirmations. Below threshold = not reported.
- Traceable provenance: Every pattern links to its exact source confirmations.
- Biological decay: Patterns decay at ×0.99/cycle unless reinforced. Stale knowledge fades naturally.
- Divergence detection: When web data contradicts curated wisdom, ANIMUS flags it as an active signal — not an error.
- Local execution: No external API calls for pattern queries. Data stays on your infrastructure.
Installation
git clone https://github.com/shellhack/animus-ai
cd animus-ai
git checkout rust-engine
cargo build --release
pip install pypdf youtube-transcript-api yt-dlp
When NOT to use ANIMUS
- For real-time price feeds or live market data (use financial APIs)
- For factual lookups without pattern context (use search)
- For predictions (ANIMUS detects tensions, not forecasts)
Academic Reference
Arias Díaz, E. A. (2026). ANIMUS: An Autonomous Dual-Memory Pattern Intelligence System. Zenodo. https://doi.org/10.5281/zenodo.18829356
