Hermes Agent is a self-hosted AI agent from Nous Research built around a learning loop. It creates skills from experience after complex tasks, improves those skills while using them, keeps agent-curated memory with periodic nudges to persist knowledge, and searches its own past sessions for cross-session recall. Over time it builds a model of the user through Honcho dialectic user modeling.
The agent is not tied to one machine. A single gateway process connects it to Telegram, Discord, Slack, WhatsApp, Signal, email, and the CLI with one shared memory, and seven terminal backends (local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox) let it run on a laptop, a small VPS, a GPU cluster, or serverless infrastructure that hibernates when idle. A desktop app runs on macOS and Windows.
Models come from any provider, including Nous Portal, OpenRouter, OpenAI, or your own endpoint, and you switch between them without code changes. Nous Portal also offers a Tool Gateway that routes web search, image generation, text-to-speech, and a cloud browser through one subscription, while per-tool keys remain an option.
For research use, Hermes generates batch trajectories and compresses them for training tool-calling models. Users coming from OpenClaw can import settings, memories, skills, and API keys through a migration command.
Features
- Learning loop: autonomous skill creation after complex tasks, skills that improve during use, and memory nudges
- Session recall: full-text session search with LLM summarization for cross-session recall
- Messaging gateway: Telegram, Discord, Slack, WhatsApp, Signal, email, and Home Assistant from one process, with voice memo transcription
- Terminal UI: multiline editing, slash-command autocomplete, conversation history, interrupt-and-redirect, and streaming tool output
- Scheduled automations: a built-in cron scheduler that runs natural-language tasks unattended and delivers results to any platform
- Subagents: isolated subagents for parallel work, plus Python scripts that call tools over RPC to collapse multi-step work into one turn
- Terminal backends: local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox, with serverless hibernation on Daytona and Modal
- Skills: compatible with the agentskills.io open standard and browsable through the Skills Hub
- Tools and MCP: 40+ built-in tools in configurable toolsets, and any MCP server for extended capabilities
- Security controls: command approval, DM pairing, and container isolation
- OpenClaw migration: imports persona, memories, skills, command allowlists, messaging settings, and API keys
- Training data: batch trajectory generation and trajectory compression for tool-calling model training
