Hermes is a persistent, self-hosted, and open-source AI agent platform designed to grow more effective over time. Unlike session-scoped chat assistants, Hermes runs as a continuous server process on hardware you control, ensuring that your data remains private and your agent's knowledge persists across all projects and sessions. Created by Nous Research, it addresses the limitations of standard assistants by providing a unified environment that connects memory, autonomous scheduling, and multi-surface communication into one cohesive system. By storing information as locally editable markdown files in a dedicated directory, it allows users to inspect, curate, or delete information at any time, effectively functioning as a long-term memory layer for your digital workflow. The platform is designed to be model-agnostic, enabling users to swap between various providers like OpenAI, Anthropic, or local open-source models based on specific needs such as cost, privacy, or performance.

Functionality-wise, Hermes acts as an autonomous assistant capable of executing tasks, managing files, writing its own reusable skills, and interacting with you through various messaging platforms. It treats the environment as an ongoing, evolving partnership where the agent learns from previous tasks and adapts its procedures without needing manual configuration for every new project.

Some of the key features are:

• Persistent Memory: A layered storage system that captures user profiles, skills, and session history in editable markdown files, ensuring context survives reboots and model swaps.
• Autonomous Scheduling: An integrated cron scheduler that runs on your server, allowing the agent to execute tasks like monitoring websites, reviewing PRs, or sending updates while you are offline.
• Multi-Surface Access: A unified agent identity accessible from platforms like Telegram, Discord, Slack, WhatsApp, and browser interfaces, allowing you to switch surfaces mid-conversation.
• Self-Improving Skills: The ability for the agent to write its own reusable skill modules from experience, which can be shared or audited via open standards.
• Extensive Tool Integration: Over 47 built-in tools covering web search, browser automation, code execution, vision analysis, and image generation.
• MCP Integration: Full support for Model Context Protocol, enabling connections to any MCP server and allowing Hermes to serve as an MCP host itself.
• Robust Security: A seven-layer defense model featuring user allowlists, dangerous command confirmation, Docker isolation, and input sanitization.
• Voice Mode: Real-time voice interaction capabilities across CLI and messaging platforms using high-performance local or cloud-based whisper models.

Hermes operates as a persistent daemon on your infrastructure. You install it via a script on a server, desktop, or even a Raspberry Pi, configure your chosen LLM provider, and then interact with it via CLI or a supported web interface. Once running, it monitors its scheduled tasks, waits for incoming messages, and maintains its memory files in real-time. Because it is self-hosted, all processing happens locally, providing full control over the agent's capability and data privacy.

Some common use cases include:

• Solo Development: Using the agent to manage project context, document conventions, and automate recurring refactoring or testing tasks.
• Shared Team Environments: Providing multiple team members with access to a capable agent that maintains shared project history and skill sets.
• Automation-Heavy Workflows: Managing daily reports, monitoring price changes, or handling incoming pull requests without needing active human oversight.
• Privacy-First Research: Conducting web research and data synthesis while keeping all intermediate data and conversation logs stored securely on private hardware.