What Hermes Agent can do that OpenClaw can't — with runnable demos.
Part of the Nous Research Hermes Agent Hackathon.
See FEATURE_DIFF.md for the full comparison. Key exclusive features:
| Feature | OpenClaw | Hermes |
|---|---|---|
| Discord Voice Channel (join VC, listen, speak) | ❌ | ✅ |
RL Training from chat (rl_* tools) |
❌ | ✅ |
| Agent-created skills (procedural memory) | ❌ | ✅ |
| Cross-session search (FTS5 + LLM summarisation) | ❌ | ✅ |
| Multiple terminal backends (Docker, Modal, SSH) | ❌ | ✅ |
execute_code (programmatic tool calling) |
❌ | ✅ |
| Honcho user modeling | ❌ | ✅ |
| Demo | File | What it shows |
|---|---|---|
| A — Discord Voice Channel | demos/demo_a_voice_vc.md |
Bot joins VC, listens, speaks back |
| B — RL Training from Telegram | demos/demo_b_rl_from_chat.md |
Launch training run from a chat message |
| C — Agent Creates a Skill | demos/demo_c_skill_creation.py |
Agent writes its own procedural memory |
| D — Cross-Session Search | demos/demo_d_session_search.md |
Find past conversations via FTS5 |
pip install git+https://github.com/NousResearch/hermes-agent.git
# Run the skill creation demo
python demos/demo_c_skill_creation.py- hermes-skill-distillation — RealWorldTaskEnv for Atropos RL training