Hermes Agent Documentation — Nous Research's Self-Improving Open-Source Agent
by Nous Research
Official docs for running, extending and hacking on Hermes Agent, the MIT-licensed agent that writes and refines its own skills.
Overview
Hermes Agent is an autonomous agent from Nous Research whose built-in learning loop creates skills from experience, improves them during use and persists knowledge across sessions. The official documentation is split into Getting Started (installation, platform support, quickstart and a learning path), Using Hermes (CLI, configuration, tools, memory, skills, MCP, voice mode, personality, context files, security), Messaging Platforms (20+ adapters including Telegram, Discord, Slack, WhatsApp, Signal and Teams), Integrations and Guides, and Developer Resources (architecture, contributing, CLI reference, FAQ). The learning path sorts readers into three tiers: beginner (about 1 hour, install through configuration), intermediate (2 to 3 hours, sessions, messaging, tools, skills, memory and cron) and advanced (4 to 6 hours, architecture, adding tools, creating skills and contributing), plus seven goal-based paths such as a CLI coding assistant, multi-bot teams and RL model training. The architecture page explains how the CLI, gateway and ACP adapter feed one AIAgent loop with 70+ tools across about 28 toolsets, seven terminal backends (local, Docker, SSH, Daytona, Modal, Singularity, Vercel Sandbox), a context compressor that summarizes middle turns, Anthropic prompt caching, SQLite session storage with FTS5 search and lineage tracking, pluggable memory providers, and ShareGPT-format trajectory export.
At a Glance
- Topic
- Agentic
- Level
- All Levels
- Format
- Documentation
- Cost
- Free
- Duration
- ~1 hour beginner path, ~2-3 hours intermediate, ~4-6 hours advanced (official learning path)
- Provider
- Nous Research
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Install Hermes Agent and connect it to Nous Portal, OpenRouter, OpenAI or a custom endpoint
- ✓Configure toolsets and terminal backends such as Docker, SSH, Modal or Daytona sandboxes
- ✓Use persistent memory and agent-created skills that the agent refines across sessions
- ✓Connect MCP servers and run the agent as a Telegram, Discord or Slack bot
- ✓Schedule recurring agent tasks with cron jobs and manage sessions across platforms
- ✓Read the AIAgent loop architecture, including context compression and prompt caching
- ✓Write custom tools that self-register and package plugins for memory providers
- ✓Export agent session trajectories in ShareGPT format for reinforcement learning and training
Highlights
- •A learning path with time estimates per tier and seven goal-based reading orders, which most agent framework docs lack
- •The architecture page names the actual modules (conversation loop, tool registry, context compressor), so you can read the code alongside it
- •The NousResearch/hermes-agent repo has about 250,000 stars and 53,000 forks and was pushed October 2, 2026 (checked October 2, 2026)
- •An independent practitioner reference by Blake Crosley (updated September 24, 2026) flags the real friction points: provider authentication, side tasks that route through the main model and add cost, and Ollama's 4,096-token default context
Who It’s For
Best For
- ✓Developers who want a self-hosted, model-agnostic agent with persistent memory and skills
- ✓Teams deploying an agent behind Telegram, Discord, Slack or other messaging platforms
- ✓Engineers studying how a production agent loop handles tools, context compression and sessions
- ✓Contributors who want to add tools or generate agent trajectories for RL training
Prerequisites
- •Comfort with a terminal on Linux, macOS, WSL2 or Windows
- •An API key or OAuth account for at least one supported LLM provider
- •Python knowledge for the advanced tier (custom tools, plugins, contributing)
FAQ
What is Hermes Agent Documentation — Nous Research's Self-Improving Open-Source Agent?
The Hermes Agent documentation is Nous Research's official guide to its open-source, MIT-licensed autonomous agent. It is for developers who want a self-hosted agent with persistent memory, self-written skills, MCP and messaging bots, and for contributors who want to add tools, write skills or export trajectories for training.
Is Hermes Agent Documentation — Nous Research's Self-Improving Open-Source Agent free?
Hermes Agent Documentation — Nous Research's Self-Improving Open-Source Agent is free to access.
What level is Hermes Agent Documentation — Nous Research's Self-Improving Open-Source Agent for?
Hermes Agent Documentation — Nous Research's Self-Improving Open-Source Agent is aimed at a all levels audience. Recommended background: Comfort with a terminal on Linux, macOS, WSL2 or Windows, An API key or OAuth account for at least one supported LLM provider, Python knowledge for the advanced tier (custom tools, plugins, contributing).
How long does Hermes Agent Documentation — Nous Research's Self-Improving Open-Source Agent take?
Expect roughly ~1 hour beginner path, ~2-3 hours intermediate, ~4-6 hours advanced (official learning path). Most learners work through it at their own pace.
What will I learn from Hermes Agent Documentation — Nous Research's Self-Improving Open-Source Agent?
You'll learn: Install Hermes Agent and connect it to Nous Portal, OpenRouter, OpenAI or a custom endpoint; Configure toolsets and terminal backends such as Docker, SSH, Modal or Daytona sandboxes; Use persistent memory and agent-created skills that the agent refines across sessions; Connect MCP servers and run the agent as a Telegram, Discord or Slack bot; Schedule recurring agent tasks with cron jobs and manage sessions across platforms; Read the AIAgent loop architecture, including context compression and prompt caching; Write custom tools that self-register and package plugins for memory providers; Export agent session trajectories in ShareGPT format for reinforcement learning and training.
Topics
Sources
This page was written from 5 sources, 2 on domains other than hermes-agent.nousresearch.com.