Hermes Agent is an open-source agent built by Nous Research and published at github.com/NousResearch/hermes-agent. You install it on your own machine or server, point it at a model provider, and talk to it through a terminal CLI or TUI, the Hermes Desktop app for macOS and Windows, or a messaging bot. The agent itself costs nothing. Model providers and optional hosted services such as the Nous Portal charge separately.
Its main selling point is a built-in learning loop. When Hermes works out a multi-step workflow, recovers from errors, or gets corrected, it can save that approach as a reusable skill and patch the skill later. It also keeps two small curated memory files (agent notes and a user profile), records every session in a searchable SQLite database, and supports external memory providers for deeper user modeling. Skills follow the open agentskills.io format, and every installed skill becomes its own slash command.
Hermes is built to run wherever you need it. It runs on Linux, macOS, WSL2, native Windows and Android (Termux), and the README says it can live anywhere from a $5 VPS to a GPU cluster. Its terminal tool can execute commands locally or inside Docker, SSH, Modal, Daytona, Vercel Sandbox or Singularity backends. A single gateway process connects the same agent and memory to Telegram, Discord, Slack, WhatsApp, Signal, email and many more platforms. That gateway also runs scheduled jobs and sends their results to whichever chat you pick.
What it can do
Self-improving skills
The agent writes and patches its own SKILL.md files with the skill_manage tool. You can also install skills from hubs such as the official repo, skills.sh, GitHub taps and well-known endpoints.
Persistent memory
MEMORY.md (agent notes) and USER.md (your profile) are loaded into every session. Both are size-limited, so the agent has to keep them concise. An optional write-approval gate lets you review changes first.
Session search
Every CLI, TUI and gateway session is stored in ~/.hermes/state.db. The session_search tool can search that history with FTS5 full-text search.
Messaging gateway
One gateway process exposes the agent on more than 25 chat platforms. Access is deny-by-default and controlled with allowlists or DM pairing codes.
Scheduled tasks
A built-in cron scheduler accepts natural-language or cron-syntax schedules and delivers results to a chat or local files. You can create jobs from the CLI, with /cron, or just by asking.
Subagent delegation
The delegate_task tool starts isolated child agents with restricted toolsets for parallel work. The execute_code tool runs Python scripts that call Hermes tools.
Sandboxed terminal backends
Commands can run locally or in docker, ssh, modal, daytona, vercel_sandbox or singularity backends. You select one with terminal.backend.
MCP client and server
Hermes connects to MCP servers over stdio or HTTP, configured under mcp_servers or added with hermes mcp. With hermes mcp serve it can also act as an MCP server itself.
Web, browser and media tools
Toolsets cover web search and extraction, browser automation, vision, image generation and text-to-speech. Voice mode works in the CLI and in messaging apps.
Profiles
Profiles are fully isolated Hermes instances, each with its own config, keys, memory, skills, sessions and cron jobs. Each profile also gets its own command alias.
Editor and API integration
hermes acp brings Hermes into editors such as VS Code, Zed and JetBrains. An API server mode exposes it as an OpenAI-compatible endpoint for frontends like Open WebUI.
Checkpoints and rollback
Opt-in filesystem checkpoints (off by default) snapshot the working directory before file changes, and /rollback restores an earlier state.
Where it runs
Model providers
Hermes supports many inference routes, including subscription portals, API providers and local model servers. Availability and pricing are set by each provider.
Messaging platforms
One gateway process connects the same agent to these chat platforms, as listed in the official messaging documentation.
License and cost
Hermes Agent is MIT-licensed open source. The agent itself is free; model providers and optional hosted services charge separately. See keeping costs under control.
Reflects the official Hermes Agent docs and GitHub README as read on 2026-10-20. The latest stable release at that time was v0.22.2 (Oct 15, 2026), with nightly canary builds on top. Commands and flags change quickly between releases, so check with hermes <command> --help.
Ready to try it? Install Hermes Agent
Free and open source from Nous Research. Pick how you want to run it:
Download the Hermes Desktop app
No terminal needed. The app includes everything Hermes needs. It runs on Macs with Apple Silicon and on Windows 11.
Step by step: Mac guide · Windows guide
Install from the Terminal
Open the Terminal app, paste this command and press Enter. It runs the official installer from Nous Research.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashWhen it finishes, open a new Terminal window and type hermes to start, then hermes setup to pick your model.
Install from PowerShell
Open PowerShell (no admin rights needed), paste this command and press Enter. It runs the official Windows installer from Nous Research.
iex (irm https://hermes-agent.nousresearch.com/install.ps1)When it finishes, open a new PowerShell window and type hermes to start, then hermes setup to pick your model.
Full guide: Windows, native or WSL2 · Problems? Install or update failed
Want it running all the time, or on a server? Run Hermes Agent 24/7. Then chat with it from your phone on Telegram, Discord or WhatsApp.
Primary sources
- Hermes Agent homepage (opens in a new tab)
- NousResearch/hermes-agent on GitHub (opens in a new tab)
- Features overview (opens in a new tab)
- Skills System (opens in a new tab)
- Persistent Memory (opens in a new tab)
- Messaging Gateway (opens in a new tab)
- AI Providers (opens in a new tab)
- Installation (opens in a new tab)
- Configuration (opens in a new tab)