Last verified: Jul 31, 2026 · Hermes 0.16.0 · OpenClaw 2026.7.1-2

Two categories that get conflated

“AI agent” in 2026 covers two very different things, and mixing them up causes most bad choices:

  1. Build-your-own frameworks — libraries you code against to assemble an agent (LangGraph, CrewAI, AutoGPT-style). You own the plumbing: tool-calling loops, memory, channels. High control, high effort.
  2. Ready-to-run harnesses — a packaged agent you install, configure, and deploy as a service. It already has the reasoning loop, the channels, the skills system. You operate it, not build it. Hermes and OpenClaw live here.

AgentHub focuses on category 2 — the self-hosted, ready-to-run harness — because that’s where a deployer (not a framework developer) needs guidance, comparisons, and migration help.

The ready-to-run harnesses (this site’s anchor)

Harness Repo Language Strength
Hermes NousResearch/hermes-agent (v0.16.0) Python Self-improving skills; strong China-channel coverage (WeChat, Feishu, DingTalk, WeCom, QQ, Yuanbao)
OpenClaw openclaw/openclaw (v2026.7.1-2) Node/TS Broadest channel coverage (24+); skills marketplace; gateway-first

Both are MIT-licensed and self-hostable. See Hermes vs OpenClaw for the head-to-head.

The build-your-own frameworks (for context)

These are not harnesses — they’re libraries for building your own. Listed for orientation, not endorsement; verify current state at each repo.

If you want to build an agent from primitives → these. If you want to deploy a ready agent to your channels → Hermes/OpenClaw.

Where AgentHub fits

This site is the hub for category 2: authoritative comparisons, install/deploy guides, migration between harnesses, and (Phase 2) a cross-harness skills directory. The framework ecosystem already has LangChain’s docs + community; the harness ecosystem doesn’t have a central hub yet — that’s the gap.

A note on freshness

Hermes + OpenClaw facts here were verified 2026-07-31 against the repos + running instances. Framework entries are orientation pointers — check each repo before relying on specifics.

Sources