An agnostic cognitive architecture for LLM-based agents, providing structured state management, memory orchestration, critical reasoning, and decision execution through a portable skill implementation.
Etta is a cognitive architecture designed to provide structure, consistency, and reasoning workflows for Large Language Model (LLM) based agents.
Rather than acting as a simple prompt or instruction set, Etta defines a formal operational model composed of states, memory systems, decision layers, critical evaluation mechanisms, and runtime execution rules.
The architecture is designed to be:
- Model-agnostic
- Platform-independent
- Portable across LLM environments
- Deterministic in structure
- Extensible by design
Etta enables agents to operate through defined cognitive processes instead of relying solely on emergent model behavior.
Modern LLMs are capable of reasoning, planning, and contextual understanding, but these capabilities are often expressed inconsistently.
Most agent implementations rely on:
- Prompt engineering
- Ad-hoc workflows
- Custom orchestration logic
- Platform-specific implementations
This creates systems that are difficult to maintain, reproduce, and evolve.
Etta addresses this problem by defining a reusable cognitive framework that establishes:
- How state is managed
- How memory is used
- How decisions are evaluated
- How reasoning is validated
- How responses are generated
The result is a portable architecture that can be implemented across different environments while maintaining consistent operational behavior.
Agent behavior is organized through explicit cognitive states rather than implicit model assumptions.
Information can be retrieved, evaluated, and incorporated into decision making through defined memory workflows.
Reasoning is subject to internal validation and review mechanisms before execution.
State management, memory, decision making, and execution are treated as distinct architectural components.
The architecture is not tied to any specific language model, framework, or deployment environment.
Etta is composed of several major subsystems:
Responsible for managing operational state, transitions, and contextual execution conditions.
Provides mechanisms for memory storage, retrieval, prioritization, and utilization.
Evaluates reasoning quality, detects inconsistencies, and supports decision validation.
Coordinates available information and produces actionable outcomes.
Orchestrates execution flow across all architectural components.
Etta/
│
├── readme_en.md
│
├── docs/
│ ├── etta_concepts.md
│ ├── etta_architecture.md
│ ├── etta_design_philosophy.md
│ ├── etta_design_principles.md
│ ├── etta_roadmap.md
│ └── etta_changelog.md
│
├── examples/
│ ├── basic_conversation.md
│ ├── memory_recall.md
│ ├── critic_intervention.md
│ ├── state_transition.md
│ └── planning_workflow.md
│
├── skill/
│ ├── SKILL.md
│ └── references/
│ ├── implementation.md
│ └── schemas.md
│
└── specifications/
├── etta_architecture_spec.md
├── etta_functional_architecture_spec.md
├── etta_state_model_spec.md
├── etta_memory_spec.md
└── etta_runtime_spec.md
If you are new to Etta, read the documentation in the following order:
Start with:
docs/concepts.md
This document introduces the fundamental concepts that define the architecture.
Continue with:
docs/architecture.md
This document explains how the system components interact.
Review:
examples/
to see practical execution scenarios.
Read:
skill/SKILL.md
to understand the portable implementation layer.
Finally, explore:
skill/references/
for implementation details, contracts, and runtime specifications.
Etta is not:
- A chatbot
- A language model
- A prompt collection
- An agent framework
- A workflow engine
Instead, Etta defines the cognitive architecture that can be implemented within such systems.
The long-term objectives of Etta are:
- Consistent agent behavior
- Structured reasoning
- Explicit state management
- Reusable cognitive workflows
- Portable implementations
- Architecture-driven development
"Etta" is derived from Henrietta Swan Leavitt, whose discovery of the Cepheid period–luminosity relationship provided a systematic way to measure cosmic distances.
Just as Leavitt's work turned observations into a navigable model of the universe, Etta Cognitive OS aims to turn language-model outputs into structured cognition through state, memory, reasoning, and execution.
Read more: https://www.womenshistory.org/education-resources/biographies/henrietta-swan-leavitt
Refer to the repository license file for usage, distribution, and contribution terms.
Current Public Release:
Etta Cognitive OS v1.0
This repository contains the public specification and portable skill implementation of the Etta architecture.
Etta™ is a research project developed by BDEV as part of an ongoing initiative exploring cognitive architectures and skill-based systems for LLM agents.
Copyright © 2026 BDEV. All rights reserved.