Aug 5, 2026 · 7 min read

Last updated on Sep 16, 2026

Hermes Agent Complete Guide: Local Agents, Tools and Memory (2026)

Hermes Agent is an open-source general-purpose agent from Nous Research. It combines tools, persistent memory, reusable skills, messaging gateways, scheduled jobs, and local or hosted model providers in one runtime.

Update (September 16, 2026): Hermes Agent v0.21.3 (v2026.9.14) is the current tagged patch release checked here. Remote Desktop/Cloud gateway refresh requests now coalesce concurrent use of a rotating refresh token, preventing refresh bursts from revoking a healthy session. Long-lived processes also stop leaking duplicate state.db writer handles. The full curated feature record is deferred to v0.22.0; this is not a separate new agent launch.

For git installations, the release documents hermes update. Docker/Cloud images build from nousresearch/hermes-agent:v2026.9.14. Back up persistent state and check your deployment’s restart procedure before upgrading; these fixes do not justify deleting state.db or weakening authentication. See the self-hosted setup guide for deployment context.

If you want an AI agent that runs on your own server, talks to you via Telegram or Discord, and gets smarter the more you use it, Hermes Agent is the best option available.

What is Hermes Agent?

Hermes Agent is a self-improving AI agent built by Nous Research. It’s not just a coding agent or a chatbot. It’s a general-purpose agent that:

Creates skills from experience
Improves skills during use
Persists knowledge across sessions
Searches its own past conversations
Builds a model of who you are
Runs on your own server (not a cloud service)
Talks to you via Telegram, Discord, Slack, WhatsApp, Signal, or CLI

The key differentiator: the learning loop. Most agents start fresh every session. Hermes Agent remembers what it learned, creates reusable skills from complex tasks, and improves those skills over time.

Key specs

Spec	Value
Developer	Nous Research
Current release checked	v0.21.3 / `v2026.9.14`
License	MIT
Launch date	February 2026
Languages	Python, TypeScript
Models	Any (Nous Portal, OpenRouter, OpenAI, custom)
Platforms	macOS, Windows, Linux, Termux
Messaging	Telegram, Discord, Slack, WhatsApp, Signal
Tools	40+ built-in
Installation	One-liner script

The learning loop

This is what makes Hermes Agent different from every other agent framework:

1. Memory: Carries preferences, project context, and relevant history into new sessions. Doesn’t start cold.

2. Skills: Repeated patterns become reusable skills you can inspect, edit, or remove. Skills self-improve during use.

3. Lessons: Corrections become durable lessons that change later behavior. Workspace-scoped lessons for project-specific guidance.

4. User modeling: Uses Honcho dialectic user modeling to build a deepening understanding of who you are and how you work.

5. Session search: FTS5 session search with LLM summarization for cross-session recall. Can find and reference past conversations.

This isn’t just “save the conversation history.” It’s a genuine learning loop where the agent creates knowledge from experience and applies it later.

Installation

Linux, macOS, WSL2

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Windows (PowerShell)

iex (irm https://hermes-agent.nousresearch.com/install.ps1)

After installation

source ~/.bashrc    # or source ~/.zshrc
hermes              # start chatting

Configuration

hermes setup        # full setup wizard
hermes model        # choose your LLM provider
hermes tools        # configure enabled tools
hermes config set   # set individual config values
hermes gateway      # start messaging gateway

Using Nous Portal

If you don’t want to collect separate API keys:

hermes setup --portal

This gives you access to 300+ models and the Tool Gateway (web search, image generation, TTS, cloud browser) under one subscription.

Using other providers

hermes model openrouter:anthropic/claude-sonnet-5
hermes model openai:gpt-5.6-sol
hermes model custom:https://your-endpoint.com/v1

Messaging gateway

Hermes Agent can talk to you from multiple platforms simultaneously:

hermes gateway setup    # configure platforms
hermes gateway start    # start the gateway

Supported platforms:

Telegram
Discord
Slack
WhatsApp
Signal
Email

All from a single gateway process. Conversation continues across platforms.

Tools and capabilities

Hermes Agent includes 40+ built-in tools:

Code: Write, edit, test, debug code
Files: Read, write, search, manage files
Shell: Run commands, manage processes
Web: Search, browse, fetch web content
Git: Commit, push, pull, manage repos
Memory: Store, recall, search knowledge
Skills: Create, improve, manage skills
Cron: Schedule recurring tasks
MCP: Connect external tools

Terminal backends

Seven terminal backends for different environments:

Local (default)
Docker
SSH
Singularity
Modal (serverless)
Daytona (serverless)
Vercel Sandbox

Daytona and Modal offer serverless persistence: your agent’s environment hibernates when idle and wakes on demand.

Scheduled automations

Built-in cron scheduler with delivery to any platform:

hermes cron add "Check open PRs and summarize" --schedule "0 9 * * *"
hermes cron add "Monitor deployment health" --schedule "*/5 * * * *"
hermes cron add "Weekly code quality report" --schedule "0 9 * * 1"

Jobs deliver results to Telegram, Discord, Slack, or wherever you’re connected.

Multi-agent coordination

Hermes Agent can spawn isolated subagents for parallel work:

Independent research and implementation
Parallel workstreams that return results to the parent
Python scripts that call tools via RPC

The main conversation stays focused on the goal while specialized agents handle work in parallel.

Security

Hermes Agent can run on your own server, but prompts and tool data still leave your infrastructure when you configure a hosted model or external service. A fully local deployment requires a local model endpoint and local tools. Relevant controls include:

Command approval (deny-by-default option)
Container isolation (Docker backend)
DM pairing for messaging platforms
Blank Slate mode (tighter tool scoping)

Migration from OpenClaw

If you’re coming from OpenClaw (the other major open-source agent in 2026):

hermes claw migrate              # interactive migration
hermes claw migrate --dry-run    # preview what would be migrated

Imports: persona, memories, skills, API keys, messaging settings, TTS assets.

When to use Hermes Agent

You want a self-improving agent that learns from experience
You need to run agents on your own server (data privacy)
You want to talk to your agent via Telegram/Discord/Slack
You prefer open-source with MIT license
You want a general-purpose agent, not just coding

When to use something else

AWS ecosystem: Kiro Crew (persistent workspace)
Python framework: CrewAI or AutoGen
Interactive coding: Claude Code, Cursor, or Kiro IDE
Cloud-first: Claude Code or ChatGPT

My take

Hermes Agent is a notable open-source option when persistent memory, reusable skills, messaging, and self-hosted execution need to live in one runtime. Whether its learning loop is useful depends on how well you audit generated skills and prevent stale or unsafe behavior from becoming persistent.

Repository popularity is not a substitute for evaluating permissions, providers, tools, and deployment behavior against your own requirements. The MIT license permits commercial use, subject to the licences and terms of the models and services you connect.

The messaging integration is the other standout feature. Talking to your agent via Telegram while it works on a cloud VM is a workflow that no other agent matches. It’s not just a coding tool. It’s a general-purpose assistant that lives where you do.

For developers who want an extensible agent on their own infrastructure, Hermes Agent is worth evaluating alongside narrower coding agents and orchestration frameworks. Choose based on permissions, model support, operational burden, and an evaluation set that reflects your real workflows.

FAQ

Is Hermes Agent free?

Yes. MIT license, open source. You pay for the model API calls (through Nous Portal or your own provider), but the agent itself is free.

How is Hermes Agent different from Claude Code?

Claude Code is a coding agent locked to Anthropic models. Hermes Agent is a general-purpose agent that works with any model, runs on your own server, talks via Telegram/Discord, and has a learning loop.

Does Hermes Agent work with any model?

Yes. Nous Portal, OpenRouter, OpenAI, custom endpoints. Switch with hermes model command. No code changes needed.

Can I run Hermes Agent on a $5 VPS?

Yes. The agent runs on minimal hardware. A $5 VPS with 1GB RAM is sufficient for most use cases. GPU is not required (the agent calls external model APIs).

What is the learning loop?

The agent creates skills from experience, improves them during use, persists knowledge across sessions, and builds a user model over time. This is unique among open-source agent frameworks.

Is Hermes Agent better than CrewAI?

They serve different purposes. Hermes Agent is a self-improving general-purpose agent with messaging integration. CrewAI is a multi-agent framework for building specialized agent teams. For personal use, Hermes. For team orchestration, CrewAI.

Can I run Hermes Agent fully locally, without any cloud API?

Yes. Point Hermes at a locally served OpenAI-compatible endpoint instead of a cloud provider. See Running Hermes Agent Locally with LFM2.5-2.6B for a documented setup using Liquid AI’s LFM2.5-2.6B, a small model specifically trained to work inside Hermes Agent.