Hermes Agent Setup: A Free, Self-Improving AI CLI

Hermes Agent is a free, open source AI agent from Nous Research that writes and improves its own skills as it works, remembers context across sessions, and runs from a terminal, a $5 VPS, or straight from Telegram. It has crossed 240,000 GitHub stars and added more than 750 of them today alone, putting it near the top of GitHub's daily trending page.

\n
• Hermes installs with one command on Windows, macOS, Linux, WSL2, and even Termux on Android; the installer bundles Python, Node.js, ripgrep, and ffmpeg for you.
\n
• It runs anywhere, not just your laptop: locally, in Docker, over SSH, or on serverless infrastructure like Modal and Daytona that hibernates when idle, and you can talk to it from Telegram, Discord, Slack, WhatsApp, or Signal, not only a terminal.
\n
• A built-in learning loop lets it create and refine its own skills from experience, search its own past conversations, and build a persistent model of who you are across sessions.
\n
• It is provider-agnostic: bring your own OpenAI, Anthropic, or OpenRouter key, or skip API keys entirely with a free Nous Portal login.
\n

The exact steps, start to finish

\n
1. Step 1. Check your machine is ready.\n

   # Windows PowerShell: confirm version and free disk space\n$PSVersionTable.PSVersion\n[System.Environment]::OSVersion.Version\nGet-PSDrive C | Select-Object Used,Free

   \nHermes needs Windows 10/11, macOS, Linux, or WSL2, and roughly a few hundred MB free before its installer pulls down its own managed Python, Node, and tool binaries. No manual toolchain setup is required; the installer handles that next.
\n
2. Step 2. Run the one-line installer.\n

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

   \nThis single PowerShell command installs everything Hermes needs (uv, Python 3.11, Node.js, ripgrep, ffmpeg, and a portable Git Bash) with no admin rights required. On Linux, macOS, or WSL2 the equivalent is a shell one-liner covered in the install section below.
\n
3. Step 3. Open a new PowerShell window. The installer updates your PATH, and a fresh terminal picks that up cleanly rather than fighting with the shell session the installer ran inside.
\n
4. Step 4. Connect a model provider, free, no API key.\n

   $ hermes setup --portal

   \nThis opens a browser to portal.nousresearch.com for a free OAuth login, sets Nous Portal as your provider, and turns on its Tool Gateway (web search, image generation, TTS) in one step. No API keys to copy anywhere. If you'd rather bring your own OpenAI, Anthropic, or OpenRouter key instead, run hermes model and pick it from the interactive list.
\n
5. Step 5. Verify the install.\n

   $ hermes doctor

   \nThis checks your config, provider auth, and installed tools, and flags anything still missing before you rely on it for real work.
\n
6. Step 6. Have your first real conversation.\n

   $ hermes --tui

   \nThen ask it something concrete and easy to verify, straight from the project's own quickstart: "Summarize this repo in 5 bullets and tell me what the main entrypoint is." A working setup shows your chosen model in the welcome banner, replies without error, and can keep the conversation going for more than one turn.
\n

What is Hermes Agent and why is it trending?

Hermes Agent is Nous Research's self-improving AI agent, built in Python and MIT licensed, with more than 240,000 GitHub stars and 49,000 forks. Under the hood it is two things: a full terminal interface (a classic CLI plus a newer TUI with multiline editing, slash-command autocomplete, and streaming tool output) and a messaging gateway that connects the same agent to Telegram, Discord, Slack, WhatsApp, Signal, and email from a single process. What sets it apart from a typical wrapper around a chat model is the "closed learning loop": Hermes curates its own memory with periodic nudges to persist what it learns, creates skills automatically after complex tasks, improves those skills as it reuses them, and runs full-text search over past sessions to recall relevant history. It also supports scheduled cron automations delivered to any connected platform, and can spawn isolated subagents to parallelize multi-step work. The latest release, v0.21.0 (tagged v2026.8.31), shipped August 31, 2026.

RelatedStrix Setup: Run an AI Penetration Tester on Your Code

How do you install Hermes Agent on Windows, macOS, and Linux?

Every platform uses the same one-line-installer pattern; only the shell changes. On Windows, run the checklist's PowerShell command above, or download the Hermes Desktop installer from the project's site for a combined CLI-plus-desktop-app install. Native Windows runs Hermes without WSL: CLI, gateway, TUI, and tools all work directly, and the installer drops a portable, isolated MinGit build if you don't already have Git.

On Linux and macOS (including WSL2 and Termux on Android), the installer is a single curl command:

# Linux, macOS, WSL2, Termux\ncurl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Reload your shell afterward so the new PATH entries take effect:

source ~/.bashrc    # or: source ~/.zshrc

Then start it the same way on every platform:

$ hermes              # start chatting!

Android users on Termux should follow the project's dedicated Termux guide, since the full extras bundle pulls in voice dependencies that aren't Android-compatible; Termux installs a curated subset instead.

RelatedVoiceStudio Setup: Clone Voices Locally, No Cloud

How does Hermes Agent compare to other AI agent tools?

Hermes even ships a one-command migration path for people coming from OpenClaw (hermes claw migrate), importing its persona file, memories, skills, and API keys automatically. Against a closed chat app, the tradeoff is the usual local-agent one: more setup, in exchange for owning the runtime, the model choice, and where your data actually goes.

What are the gotchas before you rely on it?

A few things worth knowing before you build a routine around it. First, Hermes runs real shell commands on your machine when it uses its terminal tool; for anything beyond casual local use, set up command approval or run it sandboxed with hermes config set terminal.backend docker. Second, its GitHub issue tracker is large and active (north of 39,000 open issues against 240,000+ stars), a sign of a genuinely fast-moving project more than a broken one, but still worth checking before you lean on an edge-case feature. Third, some antivirus tools (Windows Defender, Bitdefender) flag the bundled uv.exe Python package manager as a false positive; the project's README documents an attestation-based way to verify the binary and whitelist the folder if that happens to you. Fourth, if you point Hermes at a local or self-hosted model instead of a hosted provider, it requires at least 64,000 tokens of context, smaller windows get rejected at startup because multi-step tool-calling needs that much working memory.

Our take

The individual pieces of Hermes Agent, a terminal chat client, a messaging bot, cron jobs, MCP support, aren't new ideas on their own. What makes it worth the ten-minute setup is that Nous Research bundled all of them behind one install command and one config file, then added a memory and skills layer that actually tries to get better at your specific workflow instead of resetting every session. The free Nous Portal path removes the usual first-hour friction of collecting five API keys before you can even say hello, and the OpenClaw migration path suggests Nous is actively courting people already running a similar setup elsewhere. The real test is the one the project's own docs are honest about: if it can't hold a normal conversation cleanly, don't bother layering on the gateway, cron, or skills yet. Get one clean chat working first. Everything else here builds on that.

Original analysis by GenZTech. Tool documentation: NousResearch/hermes-agent on GitHub.