Cognitive AI Agent from Scratch (No LangChain, No LangGraph) — Memory+ Perception+ Decision+ Action

I built a fully autonomous cognitive agent in Python — with zero agent frameworks. No LangChain. No LangGraph. Just clean architecture, typed contracts, and four layers that think like a brain.

🧠 ARCHITECTURE: Memory → Perception → Decision → Action

Each layer has a single responsibility:
• Memory — Persists facts across sessions using keyword search + LLM classification
• Perception — Decomposes queries into trackable goals, marks them done as evidence arrives
• Decision — Tags reasoning type (LOOKUP / SEARCH / FETCH / COMPUTE / SYNTHESISE) and picks the right tool
• Action — Dispatches MCP tools and stores large results as session artifacts
🔗 RESOURCES
GitHub: https://github.com/Prerit-112/Cognitive-Agentic-Bot
LLM Gateway V3 docs: /llm_gateway/llm_gatewayV3/README.md
Tavily API: https://tavily.com

If this helped you, like and subscribe — more agentic AI builds coming soon 🚀

#ai #agenticai #architecture #learning #cognitivescience #pydantic #mcp

Видео Cognitive AI Agent from Scratch (No LangChain, No LangGraph) — Memory+ Perception+ Decision+ Action канала Prerit Saxena