Agent Learning Hub

What Agent Learning Hub does

Agent Learning Hub is a learning roadmap for people who want to build useful, reliable agents instead of collecting random links. It is kept as a single README by a Datawhale member and gathers community talks, official blogs, papers, open-source projects and engineering experience into a to-do list that readers tick off. The heading and some sections are in English, and most of the body is in Chinese.

The roadmap runs in stages: Stage 0 on what an agent is, Stage 1 on building a minimal agent loop with tool calls, Stage 2 on tool use, RAG and memory with reference projects such as GPT Researcher, AnythingLLM, mem0 and Letta, and Stage 3 on studying one modern agent harness. A What To Learn Now table ranks coding agents like Claude Code and Codex first, agent harness engineering second, and OpenClaw and Hermes-style personal agents third, followed by skills, MCP, A2A and ACP, and then evaluation and safety. It also includes a Project Ladder of runnable projects and a list of curated resources.

Key features

• Stage-by-stage to-do list from agent basics to harness engineering
• Priority table of what to learn now
• Reference open-source projects for each stage, such as GPT Researcher and mem0
• Project Ladder with a runnable project at each level
• Curated official docs and papers

When to use it

• Planning a self-study path from agent basics to a working project
• Deciding which personal-agent systems, such as Hermes-style agents, to study
• Picking open-source reference projects for RAG and memory work

Who it is for: Developers new to AI agents, and LLM application developers who want to move into agent loops, tool calling, evaluation and engineering.

How it fits with Hermes Agent

Hermes Agent is not the subject of the repository. The README ranks OpenClaw and Hermes-style personal agents as a third-priority area, describing them as long-running, local-first, cross-app agents with memory, skills and messaging entry points.

FAQ