Hermes Agent provides an open-source, autonomous AI agent designed to be a persistent personal assistant. It integrates deeply with messaging platforms and the terminal, learning user projects and building its own capabilities over time. Ideal for terminal-centric users, researchers, and power users, Hermes aims to become more capable and personalized the longer it runs, automating tasks and providing assistance across various communication channels.

How It Works

Hermes Agent operates on a foundation of persistent memory, allowing it to retain context and learn from past interactions. When it solves a problem, it automatically generates a "skill" document, enhancing its future performance and enabling knowledge sharing via a Skills Hub. It supports a flexible inference backend, allowing users to switch between Nous Portal, OpenRouter, or custom VLLM/SGLang endpoints without code modifications. The core user interface is a full terminal-based TUI, prioritizing efficiency for command-line users.

Quick Start & Requirements

Installation on Linux/macOS is via a shell script: curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash. The installer provisions Python 3.11 (if needed) using uv and sets up dependencies. Windows users can use a PowerShell script. Key requirements include Git; Node.js is optional for browser automation. Configuration involves setting API keys, typically in ~/.hermes/.env.

Highlighted Details

• Terminal-First TUI: Offers multiline editing, slash-command autocomplete, conversation history, and interrupt/redirect capabilities.
• Multi-Platform Gateway: Connects to Telegram, Discord, Slack, and WhatsApp, enabling cross-platform message mirroring and a unified interface.
• Flexible Sandboxing: Supports local, Docker, SSH, Singularity, and Modal terminal backends for secure command execution, with persistent workspaces.
• Autonomous Skills: Agents learn and create skills (documentation) for reuse, following the agentskills.io standard.
• Scheduled Automations: Includes a built-in cron scheduler for running tasks unattended across configured platforms.
• Research & RL: Features a batch runner for data generation and Atropos RL environments for training tool-calling models.

Maintenance & Community

Developed by Nous Research. The README does not explicitly list community channels like Discord or Slack, nor does it detail specific maintainers or sponsorships.

Licensing & Compatibility

The project is released under the MIT License, permitting broad use, including commercial applications.

Limitations & Caveats

The RL training integration is noted as "In Development" and not yet functional. WhatsApp sessions may require re-pairing if the protocol is updated. For maximum security, running commands via sandboxed terminal backends (Docker, SSH, etc.) is recommended over the default local execution. Edge TTS requires ffmpeg for optimal voice bubble integration on Telegram.