cognitive-agent 0.1.1

A lightweight, reasoning-driven cognitive agent framework with planning, acting, reflection, and usage tracking.

🧠 Cognitive Agent

Cognitive Agent is a lightweight, model-agnostic framework for building intelligent, reasoning-driven agents.
It supports planning, acting, reflection, tool use, and supervised replanning — without requiring LangChain or any external orchestration library.

This package is designed for developers who want full control and transparency over the agent reasoning pipeline.

---

🚀 Features

• Model-Agnostic Core – Works with any callable LLM (OpenAI, Gemini, Anthropic, Ollama, or your custom model).
• 3-Tier Memory System
  • short_term – recent context
  • working – temporary per-run state
  • long_term – persistent summary storage
• Supervisor Loop – Automatic replanning if confidence is low.
• Tool Integration – Register and dynamically select from a set of external functions.
• Knowledge Graph Extraction – Builds a simple knowledge graph from text.
• Streaming Support – Stream model thinking and responses with callbacks.
• Metrics & Token Accounting – Aggregates usage, time, and step statistics.
• Plugin-Ready Hooks – Customizable planner and reflector logic.

---

🧩 Installation

pip install cognitive-agent