hermes-agent: a self-improving AI agent with a closed learning loop and cross-platform messaging integration

What it solves

Hermes Agent is a self-improving AI agent designed to move beyond static interactions by implementing a closed learning loop. It solves the problem of AI agents forgetting user preferences, failing to retain complex task knowledge, and being locked into specific hardware or messaging platforms.

How it works

The agent utilizes a built-in learning loop to create and refine skills from experience and maintain a deepening model of the user across sessions. It is model-agnostic, allowing users to switch between providers (like OpenAI, OpenRouter, or Nous Portal) without code changes. The system can be deployed across various backends—including local machines, Docker, SSH, and serverless infrastructure like Modal—and can be accessed via a terminal interface (TUI) or a messaging gateway that connects to platforms like Telegram, Discord, and Slack.

Who it’s for

It is built for power users, developers, and AI researchers who want a persistent, cross-platform agent that can automate tasks via a cron scheduler, spawn subagents for parallel work, and maintain long-term memory of user interactions.

Highlights

• Closed Learning Loop: Automatically creates and improves skills based on experience and uses FTS5 session search for cross-session recall.
• Multi-Platform Access: A single gateway process allows interaction via Telegram, Discord, Slack, WhatsApp, Signal, and CLI.
• Flexible Deployment: Supports six terminal backends, including serverless options that hibernate when idle to reduce costs.
• Extensible Tooling: Integrates with MCP (Model Context Protocol) servers and includes a built-in cron scheduler for natural language automations.
• Research Capabilities: Supports batch trajectory generation and compression for training future tool-calling models.