Hermes Agent Research Notes

1. Project Overview

Hermes Agent is an open-source, self-learning AI agent framework developed by Nous Research.

Core philosophy: an agent should grow alongside its user — through a built-in learning loop, it creates skills from experience and continuously improves. The more you use it, the better it gets.

2. Core Features

2.1 Self-Learning Feedback Loop

• Automatically creates reusable Skill documents after completing complex tasks
• Skills self-iterate and improve through usage
• Built-in FTS5 full-text search + LLM summarization for cross-session memory recall
• Honcho-based user modeling to understand who you are

2.2 Multi-Platform Integration

A single Gateway process covers: Telegram, Discord, Slack, WhatsApp, Signal, Email. Supports voice memo transcription with continuous cross-platform conversations.

2.3 Terminal Interface

Full TUI: multi-line editing, slash command completion, conversation history, interrupt redirection, and streaming tool output.

2.4 Model-Agnostic

Supports Nous Portal, OpenRouter (200+ models), OpenAI, Anthropic, Hugging Face, Xiaomi MiMo, and more. Switch with hermes model — zero code changes required.

2.5 Scheduled Tasks

Built-in Cron scheduler. Define scheduled tasks in natural language (daily digests, backups, audits) and results are automatically delivered to any platform.

2.6 Parallel Sub-Agents

Spawn isolated sub-agents for parallel workflows. Supports Python scripts that call tools via RPC, compressing multi-step pipelines into single-turn operations with zero context overhead.

2.7 Flexible Deployment

6 terminal backends: Local, Docker, SSH, Daytona, Singularity, Modal. Serverless on-demand wake-up keeps idle costs near zero.

3. Quick Start

1
2
3
4
5
6
7
8
9
10

# Install (supports Linux / macOS / WSL2 / Termux)
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

# Start
source ~/.bashrc
hermes              # Start a conversation
hermes model        # Select a model
hermes tools        # Configure tools
hermes gateway      # Start the message gateway
hermes setup        # Full setup wizard

4. Comparison with OpenClaw

OpenClaw (formerly Clawdbot/MoltBot) was released in January 2026 by Austrian engineer Peter Steinberger, and is the hottest open-source agent project of 2026 (200k+ Stars). Hermes has a clear lineage connection — it even ships a built-in OpenClaw migration tool (hermes claw migrate).

Summary: OpenClaw has a more mature ecosystem and a larger community — a better fit for users who need autonomous execution out of the box. Hermes is lighter and emphasizes a “the more you use it, the better it knows you” self-learning mechanism, making it ideal for users who want an agent that’s a long-term companion and continuously adapts to their habits. Migration paths exist between the two, so you can switch as needed.

5. Comparison with Other Tools

6. Assessment

Strengths: Unique self-learning mechanism, model-agnostic, broad platform coverage, flexible deployment, active community.

Limitations: The project is relatively new (only 2 months old), and API stability remains to be seen. Compared to mature tools like Claude Code, the ecosystem and plugin count still have room to grow.

Best Use Case: When you want a long-running personal agent that continuously learns your preferences — especially for cross-platform scenarios (Telegram, WeChat, etc.).

Sources: Hermes GitHub | Hermes Official Docs | OpenClaw GitHub | MIT Technology Review China