LangChain
The layer under most agents. Models, tools, prompts and messages in one interface.
LangGraph
Build agents as graphs. Loops, memory, human approval and full control of the flow.
Deep Agents
Agents that plan with to-do lists, keep notes in files and hand work to subagents.
Model Context Protocol
Not a framework. The standard that lets any agent plug into any tool or data source.
DeepEval
Unit tests for what your LLM app says: metrics with a Groq judge, RAG and agent scores, and a command that fails the build.
RAGAS
Scores a RAG bot's answer and its retrieval separately, so you know which half to fix.
NeMo Guardrails
Rails written as files between your app and its model, in a language of their own.
Claude Code
Anthropic's terminal agent. Reads your repo, edits it, runs your commands.
LangMem
Long-term memory for LangGraph agents: extraction, consolidation and retrieval.
LLM Fundamentals
Tokens, temperature, prompts, evals, context, cost and speed, on a real model you run.
LiteLLM
One interface and one proxy in front of a hundred model providers.
LlamaIndex
Agents that reason over your documents. Strong on retrieval and indexing.
OpenAI Agents SDK
OpenAI's own agent loop, with handoffs, guardrails and tracing built in.
Promptfoo
A config file and one command that runs your prompts and says which answers broke.
Pydantic AI
Type safe agents. If you already trust Pydantic for validation, this feels like home.
Claude Agent SDK
Claude Code as a library. Tools, permissions, hooks, sessions, skills and plugins, with every example running keyless.
CrewAI
Give each agent a role, a goal and a backstory, then let the crew work the task.
Langfuse
Open-source tracing, prompt management and evals you can host yourself.
Agent Governance Toolkit
Policy checks between an agent deciding to call a tool and the tool running.