15.1 Birth & the "Growing Agent" Philosophy
☤ "The agent that grows with you." — Hermes Agent's self-description
From OpenClaw's Cousin
OpenClaw puts the Agent in chat apps, but every capability must be maintained by humans. Hermes took another road: let the Agent write its own skills. Hermes is published by Nous Research, positioned as "OpenClaw's successor + the autonomous-evolution branch" — to this day it offers hermes claw migrate for OpenClaw users.
| Project | Core strategy | Skills come from |
|---|---|---|
| OpenClaw | "humans write skills" | people |
| Hermes | "Agent writes + iterates skills" | people + Agent |
| OpenCode / Codex | "model + built-in tools" | vendor |
Nous Research & the "Hermes" Name
Hermes comes from Nous Research (a non-profit known for open models and open Agents). The name has two layers: mythology (Hermes the messenger god, connecting worlds) and brand continuity (their model series is also named Hermes).
The 4-Layer Design Philosophy
- Self-Learning — after a task, distill the trajectory into a Skill (
SKILL.md); - Continuous Self-Improvement — not just create, but update skills offline (DSPy/GEPA);
- Active Reflection — a Nudge Engine periodically asks "what did I learn worth persisting?";
- Cross-platform persona — same human, one persona across channels.
Why Self-Evolution Matters
| Horizon | Without self-evolution | With self-evolution |
|---|---|---|
| Day 1 | user configures everything | default + self-created skills |
| Day 30 | same as day 1 (+memory data) | a dozen user-specific skills |
| Day 90 | still teaching it | predicts user preferences |
| Year 1 | like a fresh install | each user has "their own version" |
📌 Self-evolution produces compounding returns over long-term use — a property ordinary Agents don't have.
Hermes vs OpenClaw (Precise Boundary)
| Dimension | OpenClaw | Hermes |
|---|---|---|
| License | MIT | MIT |
| Author | Peter Steinberger / community | Nous Research |
| Focus | multi-channel + skill ecosystem | self-evolving skills + memory |
| Skill source | human only | human + auto + auto-iterate |
| Memory | SQLite long-term | three-layer |
| User modeling | none | Honcho dialectic |
| Migration | — | hermes claw migrate |
Verifiable Facts (per main branch)
Repository NousResearch/hermes-agent; MIT license; Python core + TS gateway; one-liner install; docs at hermes-agent.nousresearch.com/docs/; compatible with the agentskills.io open standard; 15+ messaging platforms; 200+ model backends; 6 execution backends (local/Docker/SSH/Singularity/Modal/Daytona); hermes claw migrate for OpenClaw.
Section Summary
| Topic | Key point |
|---|---|
| Positioning | self-evolving personal Agent; skills auto-created & auto-iterated |
| Author | Nous Research |
| Philosophy | self-learn / self-improve / active reflection / cross-platform persona |
| vs OpenClaw | same lineage + self-evolution branch; hermes claw migrate |
| Core differentiator | "Skill auto-creation + auto-iteration" (see 15.4) |
Next section: 15.2 Installation & Migration (from OpenClaw)