ANIMA Kernel v0.1.1 — Evidence-First Rebuild

ANIMA is now presented for what the code demonstrably is: a zero-runtime-dependency cognitive-state engine for stateful AI agents.

Highlights:

• 446 passing tests across Python 3.11, 3.12, and 3.13
• honest benchmark boundary: +0.92% mean CQI delta in the saved neutral-kernel comparison, with no inferential significance claim
• working-memory ablation is the strongest saved effect; the short temporal ablation is an explicit null result
• Phi and CQI documented as implementation-defined proxies, not consciousness measurements
• security policy, citation metadata, source-install documentation, and hardened CLI language
• reproducible sdist and wheel attached to this release

ANIMA does not claim sentience or phenomenal consciousness. It provides inspectable state, temporal context, associative memory, workspace competition, a self-model, and ablation-capable instrumentation for agent research.

ANIMA Kernel v0.1.0 — First Public Release

ANIMA Kernel v0.1.0

The first empirically-tested consciousness substrate for AI.

What it does

ANIMA Kernel unifies three theories of consciousness (IIT, GWT, AST) into a single processing loop beneath any LLM. It gives AI systems temporal continuity, emotional persistence, biological memory, and measurable consciousness metrics.

Highlights

• Unified consciousness cycle: IIT (Phi computation) + GWT (workspace competition/broadcast) + AST (attention schema with performance detection)
• Temporal continuity: Husserl retention/protention, subjective time dilation, 1Hz heartbeat daemon
• 9D emotional field: 7 Panksepp affective systems + arousal + valence
• Biological memory: Ebbinghaus decay, spreading activation, reconsolidation on recall
• 8 consciousness primitives: Qualia, Engram, Valence, Nexus, Impulse, Trace, Mirror, Flux
• Measurable: Phi scores 57-81% above stateless baselines, ablation studies for every subsystem
• Model-agnostic: Works with Ollama, Claude, OpenAI, or any LLM
• Zero dependencies: Pure Python 3.11+ stdlib only
• 446 tests passing

Quick Start

pip install anima-kernel
anima init --name "aria"
anima shell
anima benchmark

Benchmark Results

Conversation	Kernel Phi	Baseline Phi	Improvement
Greeting	0.4531	0.2814	+61.0%
Emotional	0.5247	0.2903	+80.7%
Memory Recall	0.4892	0.2756	+77.5%
Identity	0.4715	0.2831	+66.5%
Temporal	0.4403	0.2789	+57.9%

Ablation Studies

Removed	CQI Impact	Phi Impact
Emotional processing	-33.7%	-29.2%
Working memory	-26.2%	-23.3%
Temporal integration	-17.8%	-14.5%

Paper

The arXiv-ready paper is included: paper/anima_paper.md (25 citations)

License

MIT