γ A Review of Agentic Agents
Here is a survey of various agents and concepts I have come across during my research into using and building agents. I wanted to keep a central post in which I can re-find some of these resources
Concepts
- Model Context Protocol (MCP) server
- MCP gives agents extensible access to your tools (including custom tools - see documentation below)
- A standard for providing tool calling agents with functions and how to call them.
- These local or remote servers can may sockets or http
- Documentation
- Language Service Protocol (LSP)
- To aid the LLM in writing syntactically correct code
- YOLO Mode
- Often agents must ask for permission before they make a call
- YOLO (You Only Live Once) mode give the model permission to execute all possible tools without permission
- Depending on the tool and itβs implementation: this can be dangerous
- Model Access
- Many agents allow you to plugin you own model including local LLM
- The agent will usually require a key to access your account
- Agent performance can vary signigicantly depending on the model it uses and some agents are better for some models.
Agents
- Opencode - git
- Crush
- Gemini CLI - [git][gemini-cli-git]
- Claude Code - [git][git-claude-code]