Direct
Return a tool, label, score, field, or abstention without an autoregressive decoding loop.
ARCANE documentation
Build adaptive System 1 AI.
Augmented Reconstruction of Consciousness through Artificial Neural Evolution is a library for models that make fast, grounded decisions using adaptive inference, neural memory, and biologically inspired computation.
Purpose
What System 1 means in ARCANE
System 1 is an engineering profile for fast, automatic decisions, not a claim about human consciousness. An ARCANE System 1 model produces a bounded output directly, reports uncertainty, can recall relevant experience, and adapts through explicit state or fast-weight mechanisms.
Adaptive
Align state or update bounded fast weights while processing the current input.
Experience-guided
Retrieve prior decisions and use them as evidence for the current task.
Neuromimetic
Compose feedback, spiking, plasticity, and homeostatic mechanisms around the task.
Choose your path
Start with the outcome you need. Each path includes the concepts, implementation details, and working examples for that job.
Automate decisions
Use ARC 1 for tool selection, typed extraction, classification, and fast request routing.
Explore ARC 1Build adaptive models
Compose recurrent, spiking, plastic, and homeostatic mechanisms around the behavior your system needs.
Browse layersRefine at inference time
Use resonance and bounded adaptation to update state for the input currently being processed.
Learn the workflowAdd memory and grounding
Retrieve decided examples with Hippocampus or build evidence-linked reasoning with Graph of Thought.
Explore memoryDesign principles
What ARCANE gives you
- Direct outputs
- Return a tool call, label, field, score, embedding, or prediction instead of parsing generated prose.
- Controlled adaptation
- Separate learned weights, transient resonance state, and optional bounded updates.
- Memory with evidence
- Recall the examples behind a decision and add or remove experience without retraining.
- Observable behavior
- Measure accuracy, calibration, latency, state change, and adaptation cost directly.
Perceive. Recall. Adapt. Act.
01
Perceive
Encode an input, schema, sequence, or retrieved example into a shared representation.
02
Recall
Retrieve relevant decisions and examples that can guide the current response.
03
Adapt
Align internal state or bounded fast weights according to the current input.
04
Act
Return a grounded decision or pass the refined representation to the next model stage.
Open source and ready to extend
ARCANE is led by Gianne P. Bacay and released under the MIT License. The source, tests, models, and implementation history are available for inspection, reuse, and contribution.