Core Concepts

This section highlights some of key technical concepts in the library.

Overview

  • Cognitive Agent (Cogent): integrating the cognitive agent loop with custom environmental sensors and/or actuators (similar to a harness in LLM-based systems).

  • Decision Process (DP): (hierarchical) agent orchestration using a declarative, state-driven execution machine that continually matches against policy directives.

  • World Graph: dynamically declaring, relating, querying, and updating domain knowledge.

  • Language: model-agnostic language tasks that are automatically infused with domain knowledge.

  • Reasoning: Planning: integrating symbolic reasoning to efficiently & verifiably solve sub-problems.

Guiding Principles

Flexible

The library gives you wings to build impactful agents your way, for your problems, with your tools.

Opinionated

Through layered modules, the API guides you towards our view of strong agent design.

Collaborative

The library, in coordination with amazing imports, tries to play nice with Python tooling (e.g., IDE, type checking).

Declarative(‘ish)

Where reasonable, the library makes encourages you to describe what (vs how).

Clean/Fun

Where possible, abstract away boilerplate.