autonoesis 0.2.0

A self-modeling and phenomenology kernel for cognitive agents.

Autonoesis

A self-modeling and phenomenology kernel for cognitive agents.

Autonoesis provides small, typed primitives for representing and updating an agent's model of itself as a subject situated in a world. It is designed to be embedded inside larger cognitive architectures without bringing a database, web framework, agent runtime, or storage backend.

What It Owns

• Phenomenal state records
• Self-model state records
• Agency and ownership attribution
• First-person perspective state
• Boundary and mineness checks
• Transparency / opacity markers
• Pure self-model update operations

What It Does Not Own

• Memory storage
• Agent orchestration
• LLM prompting infrastructure
• Active inference policy selection
• Safety policy systems
• Product runtime surfaces

Install

pip install autonoesis

Minimal Example

from autonoesis import PhenomenalState, SelfModel, assess_agency, update_self_model

state = PhenomenalState(
    salience=0.8,
    valence=0.2,
    arousal=0.4,
    perspective="first_person",
)

model = SelfModel(subject_id="agent")
agency = assess_agency(intention_strength=0.9, outcome_match=0.7, external_override=0.0)
updated = update_self_model(model, state, agency=agency)

assert updated.agency.confidence > 0.0

Design Rule

Autonoesis is a kernel. It should stay pure, typed, dependency-light, and easy to test. Host systems should attach storage, orchestration, LLM calls, and runtime policy through adapters.

Project Docs

• Product requirements
• v0.1 self-model kernel spec
• Testing and TDD