Hermes Agent - Self-Improving AI Agent Framework

Hermes Agent is a self-improving AI agent framework developed by Nous Research. It's the only agent with a built-in learning loop — it creates skills from experience, improves them during use, nudges itself to persist knowledge, searches its own past conversations, and builds a deepening model of who you are across sessions.

Core Features

🧠 Closed Learning Loop

Hermes Agent's core advantage is its closed-loop learning capability:

Automatic Skill Creation: Creates reusable skills after completing complex tasks
Self-Improving Skills: Continuously optimizes existing skills through usage
Knowledge Persistence: Periodic nudging mechanism ensures important information is recorded
Session Search: FTS5-based full-text search with LLM summarization for cross-session recall
User Modeling: Integrated Honcho dialectic user modeling

💬 Multi-Platform Messaging Gateway

Single gateway process supporting multiple messaging platforms:

Telegram - Full Bot API support
Discord - Server and DM support
Slack - Workspace integration
WhatsApp - Personal and group chats
Signal - Privacy-first messaging
Home Assistant - Smart home integration

Supports voice message transcription and cross-platform conversation continuity.

🖥️ Real Terminal Interface

Full-featured TUI (Terminal User Interface):

Multi-line editing support
Slash command autocompletion
Session history browsing
Interrupt and redirect
Streaming tool output display

🔧 Multi-Model Support

Use any model you prefer, switch anytime:

Provider	Features
Nous Portal	Official Nous Research
OpenRouter	200+ models
z.ai/GLM	Zhipu GLM
Kimi/Moonshot	Moonshot AI
MiniMax	MiniMax models
OpenAI	GPT series
Custom endpoints	Local or private models

Switch models with one command: hermes model

⏰ Scheduled Automations

Built-in Cron scheduler with delivery to any platform:

Daily reports
Nightly backups
Weekly audits

All configured in natural language, running unattended.

🚀 Parallel Subagents

Spawn isolated subagents for parallel workflows:

Complex task decomposition
Multi-threaded processing
Python scripts calling tools via RPC
Compress multi-step pipelines into zero-context-cost turns

🌐 Runs Anywhere

Six terminal backend options:

Local - Run directly on host
Docker - Containerized deployment
SSH - Remote servers
Daytona - Serverless persistence
Singularity - HPC environments
Modal - Serverless cloud

Daytona and Modal offer serverless persistence — environments hibernate when idle and wake on demand, costing nearly nothing between sessions.

Quick Start

Installation

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Supports Linux, macOS, and WSL2. The installer handles Python, Node.js, dependencies, and the hermes command automatically.

Note: Native Windows is not supported. Please install WSL2.

Configure and Launch

# Reload shell
source ~/.bashrc  # or source ~/.zshrc

# Configure LLM provider
hermes setup

# Start chatting
hermes

Common Commands

hermes              # Interactive CLI
hermes model        # Switch provider or model
hermes setup        # Re-run setup wizard
hermes gateway      # Start messaging gateway
hermes update       # Update to latest version
hermes doctor       # Diagnose issues

Skills System

Hermes Agent's skills system is compatible with the agentskills.io open standard.

Skill Creation Flow

Trigger: After completing complex tasks, Agent evaluates skill creation
Generation: Automatically generates skill files with tool call sequences
Continuous Improvement: Collects feedback and improves with each use
Version Management: Skills can be iteratively updated

Skill Usage Example

# Skill file example (simplified)
name: "web_research"
description: "Perform deep web research and summarize"
tools:
  - web_search
  - web_fetch
  - summarize
steps:
  1. Receive research topic
  2. Execute multi-source search
  3. Fetch key page content
  4. Generate structured summary

Memory System

Three-Layer Memory Architecture

┌─────────────────────────────────────┐
│         Working Memory              │
│    Current session context & state  │
├─────────────────────────────────────┤
│         Short-term Memory           │
│    Recent session key info summary  │
├─────────────────────────────────────┤
│         Long-term Memory            │
│    Persistent user model & skills   │
└─────────────────────────────────────┘

Memory Trigger Mechanisms

Periodic Nudging: Agent proactively asks whether to record important information
Task Completion: Automatically extracts reusable knowledge
Explicit User Instruction: "Remember this..."

MCP Integration

Model Context Protocol (MCP) support for connecting any MCP server to extend capabilities:

# Add MCP server
hermes mcp add <server-name> <server-url>

# List connected MCP servers
hermes mcp list

# Use MCP tools
# Call directly in conversation, Agent routes automatically

Configuration Example

Basic Config (`~/.hermes/config.yaml`)

# LLM Provider Configuration
provider:
  default: openrouter
  openrouter:
    api_key: ${OPENROUTER_API_KEY}
    model: anthropic/claude-3.5-sonnet
  
  # Fallback provider
  openai:
    api_key: ${OPENAI_API_KEY}
    model: gpt-4o

# Messaging Gateway Configuration
gateway:
  telegram:
    enabled: true
    bot_token: ${TELEGRAM_BOT_TOKEN}
  discord:
    enabled: true
    token: ${DISCORD_TOKEN}

# Memory Configuration
memory:
  auto_save: true
  nudge_interval: 300  # Remind every 5 minutes
  max_search_results: 10

# Skills Configuration
skills:
  auto_create: true
  auto_improve: true
  hub_url: https://agentskills.io

Research Features

Hermes Agent provides specialized support for AI research:

Batch Trajectory Generation: Generate large volumes of interaction trajectories for training
Atropos RL Environments: Reinforcement learning training environments
Trajectory Compression: Compress interaction trajectories for training next-generation tool-calling models

Contributing

Hermes Agent welcomes contributions!

Developer Quick Start

# Clone repository
git clone --recurse-submodules https://github.com/NousResearch/hermes-agent.git
cd hermes-agent

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Create virtual environment
uv venv .venv --python 3.11
source .venv/bin/activate

# Install dependencies
uv pip install -e ".[all,dev]"
uv pip install -e "./mini-swe-agent"

# Run tests
python -m pytest tests/ -q

Project Structure

hermes-agent/
├── src/
│   ├── agent/          # Core Agent logic
│   ├── skills/         # Skills system
│   ├── memory/         # Memory management
│   ├── gateway/        # Messaging gateway
│   └── tools/          # Toolsets
├── scripts/            # Installation and utility scripts
├── tests/              # Test suite
└── docs/               # Documentation

Resources

License

MIT License - See LICENSE

Built by Nous Research