Hermes vs OpenClaw — Feature Matrix¶

Detailed reference for the comparison tutorial. Values reflect upstream docs as of mid-2026; verify with hermes doctor and openclaw doctor on your machine.

Core platform¶

Feature	Hermes Agent	OpenClaw
Maintainer	Nous Research	OpenClaw community (orig. Peter Steinberger ecosystem)
Install	`curl -fsSL https://hermes-agent.nousresearch.com/install.sh \\| bash`	`npm install -g openclaw@latest`
Primary language	Python	TypeScript
CLI entry	`hermes`	`openclaw`
Interactive UI	`hermes --tui`	Web dashboard + channel chat
Daemon / always-on	`hermes gateway`	`openclaw onboard --install-daemon` (launchd/systemd)
Health check	`hermes doctor`	`openclaw doctor`
Config home	`~/.hermes/`	`~/.openclaw/`

Models & providers¶

Feature	Hermes Agent	OpenClaw
Cloud APIs	Anthropic, OpenAI, Nous portal, others via `hermes setup`	Anthropic, OpenAI, Google, etc. via onboarding
Local / Ollama	Supported via provider configuration	Ollama provider — e.g. `gemma4:e2b` in our guide
Model switching	`hermes model`	`openclaw.json` + onboarding
Recommended quality	Strongest latest-gen cloud model	Same — docs recommend strongest available model

Gateway & channels¶

Feature	Hermes Agent	OpenClaw
Gateway command	`hermes gateway`	`openclaw gateway` (installed as daemon)
Channel count	18+ messaging platforms (docs)	Core channels + bundled/external plugins
Named channels	Telegram, WhatsApp, Discord, Slack, …	Telegram, WhatsApp, Discord, Slack, Signal, Teams, Matrix, iMessage, Google Chat, WebChat, Zalo, …
DM security	Pairing / allowlists — run `hermes doctor` after changes	Channel-specific pairing; see Security
WebChat	Via gateway / TUI	Built-in WebChat
Mobile nodes	Apps in Hermes monorepo	macOS/iOS/Android speak + Canvas

Skills & workspace¶

Feature	Hermes Agent	OpenClaw
Skill format	`SKILL.md` + YAML frontmatter	`SKILL.md` + YAML frontmatter
Default path	`~/.hermes/skills/<name>/`	`~/.openclaw/workspace/skills/<name>/`
Shared / global skills	Ecosystem + optional dirs	`~/.openclaw/skills` with `--global` install
Registry	Community (awesome-hermes-agent)	ClawHub + `openclaw skills`
Install skill	Clone / ecosystem scripts	`openclaw skills install <id>`
Skill verification	Ecosystem maturity tags	`openclaw skills verify` (trust envelope)
Auto skill creation	Yes — learning loop writes skills from sessions	No — author or install skills
Skill curation	Curator grades/prunes (~7-day cycle, v0.12+)	Manual update via ClawHub / git
Injected prompts	Profiles, memory context	`AGENTS.md`, `SOUL.md`, `TOOLS.md`

Tools, MCP & automation¶

Feature	Hermes Agent	OpenClaw
Built-in tools	60+ (file, shell, browser, search, …)	Tool suite + session tools
MCP	First-class optional MCP servers	Supported in agent runtime
Plugins	`plugins/` + ecosystem	Channel plugins + skill plugins
Cron	`hermes cron`	Cron jobs
Webhooks	Gateway integrations	Webhooks
Email triggers	Ecosystem / custom	Gmail Pub/Sub
Remote sandboxes	Modal, Daytona, Vercel Sandbox	Docker sandbox (recommended)

Memory & learning¶

Feature	Hermes Agent	OpenClaw
Session memory	Built-in + ecosystem memory plugins	Session history tools
Long-term memory	Memory layer + honcho/hindsight/plur (ecosystem)	Workspace + session stores
Self-improvement	Core feature — skills evolve from use	Skill updates manual / ClawHub
Evaluation loop	Curator + optional self-evolution packs	Operator-driven

Multi-agent¶

Feature	Hermes Agent	OpenClaw
Native multi-agent	Profiles, delegation patterns	Multi-agent routing per docs
Ecosystem	oh-my-hermes, ACP skill, mission-control, opencode-hermes	Session tools, agent workspaces
Handoff to Codex/Claude	`hermes-agent-acp-skill`	Via agent runtime configuration

Migration & interoperability¶

Path	Command / resource
OpenClaw → Hermes (native)	`hermes claw migrate`
OpenClaw → Hermes (community)	openclaw-to-hermes
Shared pattern	Both use `SKILL.md` — port skill folders and adapt tool paths
RAG sidecar	Run Qwen Agentic RAG API; call from either skill

Security posture¶

Topic	Hermes Agent	OpenClaw
Untrusted skills	Treat ecosystem skills as untrusted; review `SKILL.md` + scripts	Docs: treat third-party skills as untrusted code
Network exposure	Keep gateway local; use pairing until trusted	Local-first gateway; caution on remote exposure
Sandboxing	Docker image, remote terminal backends	Docker sandbox strongly recommended
Audit	`hermes doctor`	`openclaw doctor`, `openclaw skills verify`

When to choose (summary)¶

Scenario	Recommendation
Personal assistant on phone chat apps, Node shop	OpenClaw
Agent research, skill evolution, Nous stack	Hermes
Local Gemma + RAG on Telegram	OpenClaw (guide)
Leaving OpenClaw, keep channels	Hermes + `hermes claw migrate`
Maximum skills marketplace (2026)	OpenClaw (ClawHub) + Hermes (awesome-hermes-agent) — pick primary runtime, borrow skills from the other