hermes-agent

Hermes Agent is a self-improving AI agent built by Nous Research. It is designed around a closed learning loop: the agent autonomously creates skills after complex tasks, refines them during real use, periodically persists knowledge, searches its own past conversations, and builds a deepening model of the user across sessions. It is compatible with the agentskills.io open standard and can use Honcho for dialectic user modeling. The agent is provider-agnostic. Users can connect Nous Portal, OpenRouter, OpenAI, their own endpoints, and many other providers, switching with `hermes model` without code changes or lock-in. It ships with a full terminal UI offering multiline editing, slash-command autocomplete, conversation history, interrupt-and-redirect, and streaming tool output. A single gateway process connects Hermes to Telegram, Discord, Slack, WhatsApp, Signal, email, and the CLI, with features like voice memo transcription and cross-platform conversation continuity. It includes a built-in cron scheduler for natural-language scheduled tasks such as daily reports and weekly audits, which can deliver results to any connected platform. The agent can also spawn isolated subagents for parallel workstreams and call tools via RPC from Python scripts, collapsing multi-step pipelines into zero-context-cost turns. Hermes runs on many terminal backends: local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox. Modal and Daytona provide serverless persistence, where the environment hibernates when idle and wakes on demand, so it can run on a small VPS or a GPU cluster rather than being tied to a laptop. Security capabilities include command approval, DM pairing, and container isolation. Installation is a one-liner for Linux, macOS, WSL2, Termux, and native Windows via PowerShell. A migration command imports settings, memories, skills, API keys, and messaging config from OpenClaw. Documentation covers CLI usage, messaging gateway setup, configuration, tools and toolsets, the skills system, memory, MCP integration, cron scheduling, context files, architecture, and contributing. The project is MIT licensed.