Hermes Agent: Free Self-Improving AI Agent (Native App Released)

Hermes Agent is a free, open-source, self-improving AI agent from Nous Research. It runs through a CLI, TUI, native desktop app, web dashboard, messaging gateway, server, VPS, or local machine.

It learns from your conversations, creates reusable skills from completed tasks, maintains long-term memory across sessions, verifies work against evidence, and operates through Telegram, Discord, Slack, WhatsApp, Signal, email, CLI, TUI, desktop, and browser-based administration surfaces.

Hermes arrived after OpenClaw had already pushed personal AI agents into mainstream developer discussion. OpenClaw proved that an agent could live outside a browser tab, connect to personal channels, remember context, and act across local tools.

Hermes falls into the same category, with a stronger emphasis on self-improving skills, model choice, remote execution, desktop access, and a built-in migration path for OpenClaw users.

The project’s GitHub growth gives Hermes more credibility than a normal new agent launch. Hermes crossed 140,000 stars in under three months and now sits above 200,000 stars, placing it among the most visible open-source AI agent projects around the web.

Read More: Automate Anything: 10 Best & Open-source AI Agents

Most AI assistants still work like short-lived chat sessions. Hermes Agent solves a different problem: persistent AI work across projects, channels, models, and execution environments.

It can remember prior work, create reusable skills, schedule recurring tasks, run tools, delegate work to subagents, and keep running beyond a single laptop session.

Hermes Agent vs OpenClaw

Hermes is the better choice when the agent itself should improve over time and operate across servers, terminals, desktop sessions, skills, web administration, and scheduled workflows.

OpenClaw remains a great choice when the main goal is a chat-first personal assistant that connects to many messaging surfaces and device workflows.

If you already use OpenClaw, you do not need to rebuild everything from scratch before trying Hermes. The migration command imports persona files, memory entries, user-created skills, approval patterns, messaging settings, API keys, TTS assets, and workspace instructions when the source setup uses supported paths and formats.

Features

• Runs as a terminal agent, TUI app, desktop app, server agent, web dashboard, or messaging gateway.
• Provides Hermes Desktop for macOS, Windows, and Linux, with streaming chat, session search, drag-and-drop files, clipboard image paste, and in-app updates.
• Connects the desktop app to a remote Hermes gateway through OAuth or username and password login.
• Supports concurrent multi-profile sessions and cross-profile session links in the desktop experience.
• Offers a browser-based administration panel for messaging channels, MCP catalog settings, credentials, webhooks, memory, gateway controls, and system diagnostics.
• Supports Telegram, Discord, Slack, WhatsApp, Signal, Email, CLI, desktop, web dashboard, and other connected surfaces.
• Keeps persistent memory across sessions through local Hermes configuration and memory files.
• Creates and improves skills after complex tasks.
• Ships a leaner default skill set and lets you browse trusted Skills Hub taps, including NVIDIA/skills.
• Searches previous sessions through full-text search and LLM summarization.
• Switches models through hermes model and fuzzy model search across desktop, web, TUI, and CLI.
• Runs named Mixture-of-Agents presets as selectable models, with each reference model’s reasoning shown before the aggregator answer streams.
• Works with Nous Portal, OpenRouter, OpenAI, Anthropic, Hugging Face, Google Gemini, Vertex AI, Kimi, MiniMax, xAI, and custom endpoints.
• Runs scheduled automations through natural-language cron workflows.
• Spawns isolated subagents for parallel workstreams and can fan out multiple background subagents while the main chat continues.
• Verifies coding work against evidence and supports /goal completion contracts.
• Creates reusable skills through /learn and exposes memory review through /journey.
• Adds coding Projects in the desktop app, including project navigation, a coding rail, review pane, git worktree management, and agent-facing project tools.
• Runs terminal work through local, Docker, SSH, Singularity, Modal, and Daytona backends.
• Supports web search, browser automation, vision, image generation, text-to-speech, terminal tools, and file operations.
• Edits existing images through image_generate when the configured image provider supports image-to-image workflows.
• Uses /undo [N] to back up recent user turns across CLI, TUI, and messaging platforms.
• Imports OpenClaw settings, memories, skills, messaging settings, and selected API keys through hermes claw migrate.

Use Cases

• Run a personal AI assistant from Telegram while the agent works on a cloud VM.
• Use the native desktop app when you want Hermes chat, sessions, files, and model selection in a GUI.
• Use the TUI when you want a terminal interface with a fuller screen layout than the classic CLI.
• Connect a laptop desktop session to a remote Hermes gateway where your API keys and compute live.
• Configure channels, MCP catalog entries, credentials, webhooks, memory, and gateway settings from the browser dashboard.
• Build daily reports, weekly audits, backups, and recurring briefings through scheduled automations.
• Maintain project memory across long development, research, content, or operations workflows.
• Delegate larger tasks into parallel subagent workstreams.
• Send several research, audit, or coding subtasks to background subagents and review the combined result later.
• Connect one assistant to multiple model providers and switch models per task.
• Run a named Mixture-of-Agents preset when you want several frontier models to compare reasoning before one answer is returned.
• Use /goal completion contracts for work where “done” should depend on tests, checks, or other evidence.
• Review what Hermes has learned through /journey before you keep, edit, or delete saved memory and skills.
• Use Docker or SSH backends when local host access creates security concerns.
• Use the Simplified Chinese desktop interface when English-first tooling slows down setup or review.
• Migrate an existing OpenClaw setup into a Hermes workflow.

HOW TO USE IT

1. Choose the right installation path

On Linux, macOS, WSL2, and Termux, install Hermes through the shell installer.

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

On native Windows, install Hermes from PowerShell.

powershell -ExecutionPolicy Bypass -c "irm https://hermes-agent.nousresearch.com/install.ps1 | iex"

If you use the Python package path, run the post-install step after installing the package.

pip install hermes-agent
hermes postinstall

For desktop use, install the native app from Hermes Desktop or add desktop support through the CLI installer.

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

If Hermes is already installed, launch the desktop app through the CLI.

hermes desktop

2. Reload your shell after installation

The installer adds Hermes to your shell profile. Reload the profile before starting the agent.

source ~/.bashrc

If you use Zsh, reload the Zsh profile.

source ~/.zshrc

3. Run the setup wizard

The setup wizard configures the base agent, providers, tools, and local settings.

hermes setup

Nous Portal gives the shortest setup path when you want models and tool gateway access under one login. The hermes portal alias runs the quick setup flow.

hermes portal

The older portal setup command remains useful when you prefer an explicit setup flag.

hermes setup --portal

Check the Portal wiring after setup.

hermes portal status

4. Pick a model provider

Hermes uses hermes model as the provider and model selection flow.

hermes model

The model picker has fuzzy search in the desktop app, web dashboard, TUI, and CLI. Hermes also treats named Mixture-of-Agents presets as selectable models through the moa provider, so an ensemble can appear beside Claude, GPT, Grok, Gemini, and other configured models. OpenRouter works well when model variety matters. Local endpoints make sense when privacy and cost control matter. Nous Portal is the better fit when you want fewer separate API keys for model access, web search, image generation, text-to-speech, and cloud browser tools.

5. Start with a basic chat

Start the classic CLI.

hermes

Start the newer TUI.

hermes --tui

Start the desktop app when you prefer a GUI for chat, files, session search, profiles, and model selection.

hermes desktop

Use a concrete first prompt that proves file and terminal access work.

Check my current directory and tell me what looks like the main project file.

6. Confirm session continuity

Hermes becomes more useful when sessions resume correctly.

hermes --continue

The short version works too.

hermes -c

7. Configure tools before adding channels

Tool permissions matter more once Hermes starts receiving messages from chat apps.

hermes tools

Docker isolation is a safer default for terminal work that should not touch the host directly.

hermes config set terminal.backend docker

SSH works when the agent should operate on a remote machine.

hermes config set terminal.backend ssh

8. Add the messaging gateway

Configure the gateway after CLI chat works.

hermes gateway setup

Start the gateway after platform setup.

hermes gateway start

Use the gateway for Telegram, Discord, Slack, WhatsApp, Signal, Email, and other supported channels. Keep public or group channels more restricted than your private terminal session.

If you run the web dashboard, use the administration panel to manage messaging channels, MCP catalog settings, credentials, webhooks, memory, and gateway controls from the browser.

9. Add skills only after the core workflow works

Skills are reusable instruction files. They help Hermes repeat multi-step workflows such as deployment, PR review, repo analysis, reporting, and automation.

hermes skills

Hermes uses a leaner default skill set than early builds, so the picker carries less bundled noise. Add skills for repeated workflows first, then expand after the agent proves stable. The /learn command can also turn a completed workflow, directory, URL, or described process into a reusable skill. If you work with NVIDIA CUDA-X, AIQ, cuOpt, or related stacks, the NVIDIA/skills tap is a trusted Skills Hub source.

10. Migrate from OpenClaw only after reviewing the plan

Hermes imports OpenClaw data through a dedicated migration command.

hermes claw migrate

Preview the migration before writing files.

hermes claw migrate --dry-run

Run a user-data migration when you want memories and skills but not secrets.

hermes claw migrate --preset user-data

Use overwrite only when you accept conflict replacement.

hermes claw migrate --overwrite

Check the migrated setup through a new session, then verify provider keys.

hermes status

Hermes Agent Cheatsheet & Quick Reference

Installation Commands

Core CLI Commands

Configuration Examples

Slash Commands

OpenClaw Migration Commands and Flags

Provider and Environment Options

Alternatives and Related Tools

• 7 Best OpenClaw Alternatives for Safe & Local AI Agents
• Most Popular Agent Skills on GitHub for Coding Agents
• Hermes Agent Documentation
• Hermes Desktop

Pros

• Free and open-source.
• MIT license.
• Strong memory system.
• Skill creation loop.
• Multi-provider support.
• Messaging app access.
• Native desktop app for macOS, Windows, and Linux.
• Remote gateway connection from the desktop app.
• Browser-based administration dashboard.
• OpenClaw migration command.
• Selectable Mixture-of-Agents presets.
• Evidence-based completion checks for coding work.
• /learn, /journey, and /goal support.
• Background subagent fan-out.

Cons

• Setup takes planning.
• Tool permissions need care.
• Local access can create risk.
• Remote gateway profiles require careful credential handling.
• Configured workflows need maintenance.
• More complex than chatbots.

FAQs

Q: Is Hermes Agent free?
A: Hermes Agent is free and open-source under the MIT license. Model usage, hosted providers, cloud machines, and paid gateway services may still cost money.

Q: Who publishes Hermes Agent?
A: Hermes Agent is published by Nous Research. The source code, releases, installation files, and issue history are available in the NousResearch/hermes-agent GitHub repository.

Q: Does Hermes Agent run on Windows?
A: Hermes Agent runs on native Windows through a PowerShell installer. WSL2 also works with the Linux installer, and the desktop app runs on Windows 10 and Windows 11.

Q: Does Hermes Agent have a desktop app?
A: Yes. Hermes Desktop is a native app for macOS, Windows, and Linux. It has chat streaming, session search, file drag and drop, clipboard image paste, model selection, profile management, and in-app updates.

Q: Does Hermes Agent have a TUI?
A: Yes. Run hermes --tui to start the terminal UI. Use the TUI when you want a more structured terminal interface than the classic interactive CLI.

Q: Can Hermes Desktop connect to a remote Hermes gateway?
A: Yes. Hermes Desktop connects to a remote Hermes gateway with OAuth or username and password login. This is useful when you want the interface on your laptop and the agent runtime on a server, hosted machine, or teammate-managed gateway.

Q: What changed in Hermes Agent v0.18.0?
A: Hermes Agent v0.18.0 added first-class Mixture-of-Agents model presets, visible reasoning from reference models, streamed aggregator answers, evidence-based work verification, /goal completion contracts, /learn, /journey, background subagent fan-out, coding Projects in the desktop app, gateway scale-to-zero, drain coordination, and Vertex AI support for Gemini.

Q: Can Hermes Agent replace OpenClaw?
A: Hermes Agent can replace OpenClaw for many long-running personal agent workflows, especially when memory, skills, model control, desktop access, and server execution matter.

Q: Can Hermes Agent import OpenClaw data?
A: Hermes Agent imports supported OpenClaw data through hermes claw migrate. The migration covers persona files, memory, user profiles, skills, approval patterns, messaging settings, selected API keys, TTS assets, and workspace instructions.

Q: Is Hermes Agent private?
A: Hermes keeps memory, configuration, and local workflows on your own machine or server when you run it locally. Prompts, code, files, and task data may leave your environment when you use external model providers, cloud browsers, web tools, hosted services, or remote gateways.

Q: Is Hermes Agent good for non-developers?
A: Hermes Desktop makes the agent more approachable, but Hermes still works best when you understand model providers, permissions, automation, and basic setup concepts. A simple hosted chatbot suits casual use better.

Q: What is the biggest limitation of Hermes Agent?
A: Hermes Agent needs careful setup before serious use. Model choice, gateway access, tool permissions, sandboxing, memory behavior, remote profiles, and scheduled automations all affect reliability and safety.

Last Updated: July 01, 2026