AI software · verified 05 Nov 2026

Hermes AI Agents

The self-improving, open-source agent from Nous Research (MIT license): persistent memory, self-written skills, a 20+ platform messaging gateway, and scheduled automations — running on your VPS, not your laptop. This hub collects verified facts, the architecture, install paths, a security briefing, and a hands-on console demo.

Official repository ↗ Official docs ↗ Directory profile

MITopen-source license

40+built-in tools

20+messaging platforms, one gateway

6terminal backends (local → serverless)

The closed learning loop

YouTelegram, WhatsApp, CLI…

Gatewayone process, 20+ platforms

Agentplans + 40+ tools

MemoryMEMORY.md · USER.md · FTS5 recall

SkillsSKILL.md · agentskills.io

After complex tasks the agent writes a reusable SKILL.md procedure; memory nudges keep MEMORY.md and USER.md curated; full-text search (FTS5) with LLM summarization recalls past sessions. Sub-agents and a cron scheduler deliver reports to any platform — unattended.

What it does

Lives where you do

Telegram, Discord, Slack, WhatsApp, Signal, Matrix, email, SMS, Teams and more — one gateway process, cross-platform conversation continuity, voice-memo transcription.

Any model, no lock-in

Nous Portal, OpenRouter, OpenAI or any OpenAI-compatible endpoint. Switch with hermes model — no code changes. Works with local endpoints too.

Scheduled automations

Built-in cron in natural language: daily reports, nightly backups, weekly audits — delivered to any platform. Results only when something changes.

Runs anywhere

Six terminal backends: local, Docker, SSH, Daytona, Modal, Vercel Sandbox (+ Singularity plugin). Serverless options hibernate when idle — a $5 VPS is enough to start.

Extensible

40+ built-in tools (web, files, terminal, browser, vision, TTS, image generation), MCP server support, and a Skills Hub compatible with the agentskills.io open standard.

Delegates & parallelizes

Spawn isolated sub-agents for parallel workstreams, each with its own conversation and terminal; programmatic tool calling collapses pipelines into single calls.

Install — three paths

Commands below are copied from the official repository and docs. Always inspect an install script before piping it to a shell, and verify the URL is github.com/NousResearch/hermes-agent.

Linux / macOS / WSL2

# inspect first, then run
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
source ~/.bashrc
hermes setup     # provider + model wizard
hermes doctor    # smoke test

Windows (native PowerShell)

iex (irm https://hermes-agent.nousresearch.com/install.ps1)
# installs under %LOCALAPPDATA%\hermes
hermes setup
hermes doctor

Docker (always-on)

docker run -d --name hermes \
  -v ~/.hermes:/opt/data \
  nousresearch/hermes-agent
# state persists in ~/.hermes;
# update = replace the image

After install: pick a provider (hermes setup --portal for the all-in-one Nous Portal route), connect a messaging platform, then run one hermes chat -q "hello" smoke test before wiring automations.

Security briefing

Read before giving it tools. Hermes ships command-approval gates, DM pairing codes, user allowlists and container isolation — use all of them. The project's own docs advise a container backend and never running the gateway as root. Publicly tracked issues include a prompt-injection gap (#27284) and an approval-bypass risk for non-shell tools such as send_message and write_file (#35357) — check their status before unattended deployment. Give agents least-privilege credentials, require approval for destructive actions, and never point a shell-capable agent at production secrets.

Try the loop — console demo

Front-end demonstration. Responses are pre-scripted and run 100% in your browser (history stays in localStorage). No model, agent, or server is connected.

Ecosystem & next steps

agentskills.io ↗ — the open skill standard and Skills Hub; hub installs pass a security scanner
OmniRoute (listed here) ↗ — route Hermes and coding tools across free model tiers
Open WebUI (listed here) ↗ — self-hosted chat front-end for your own models
AI Agents & Gateways collection ↗ — the full self-hosting stack

Guide: Self-hosted AI agents in 2026 ↗ — how the layers fit together
Hermes Agent directory profile ↗ — pricing model, license, sources, last-checked date
Official documentation ↗ — messaging gateway, tools, MCP, cron, security
Submit a tool ↗ — found something in the ecosystem we should list?

Content on this page is English-only at launch (interface is translated). Facts verified against the official repository, official docs and corroborating sources on 05 Nov 2026. This is a directory feature — WAKITRON 2 does not host, run or resell Hermes Agent.