Hermes Agent is a free and open-source personal AI agent from Nous Research that can work across the terminal, desktop app, browser, messaging channels, and remote machines while keeping memory and reusable skills between sessions. It can run tools, delegate tasks to subagents, and switch among configured model providers.
This agent is good for projects where you need persistent AI work across terminals, messaging channels, desktop sessions, and remote execution. It operates through Telegram, Discord, Slack, WhatsApp, Signal, email, the CLI, TUI, desktop app, and browser-based administration UI.
Review tool permissions, model-provider access, browser access, MCP servers, and gateway credentials before connecting Hermes Agent to real projects or chat channels.
Hermes Agent vs OpenClaw
OpenClaw and Hermes Agent overlap in messaging, automation, remote execution, persistent state, and multi-model support. Hermes centers multi-agent work on reusable profiles, Bot Mode, learned skills, CLI/TUI control, MCP management, and model-provider configuration. OpenClaw centers the experience on a personal assistant that works across messaging, native apps, paired devices, and cloud workers.
| Hermes Agent | OpenClaw 2.0 | |
|---|---|---|
| Primary focus | Persistent AI agents with memory, skills, Bot Mode, subagents, model choice, desktop/terminal access, and remote execution. | Personal AI assistant across messaging, native apps, paired devices, cloud workers, automations, and interactive sessions. |
| Multi-agent work | Bot Mode, profile-based bots, group chats, direct bot DMs, peer gateways, and steerable delegated subagents. | Subagent tasks, durable progress views, paired devices, cloud workers, and session-based task execution. |
| Scheduled work | Cron jobs, Bot routines, persistent memory, continuity between runs, per-job work directories, and script-only monitoring jobs. | Recurring automations with approval controls and delivery through the personal-assistant environment. |
| Model access | Many model providers, OAuth options, API keys, custom OpenAI-compatible endpoints, self-hosted models, and model switching. | Multiple model providers configured through the OpenClaw runtime and apps. |
| Apps and access | CLI, TUI, Hermes Desktop, web dashboard, messaging gateway, API server, and remote gateways. | Web and native chat apps, messaging channels, gateway services, and paired-device experiences. |
| Memory and skills | Persistent memory, reusable skills, /learn, /journey, profile-specific state, and Bot chats. | Persistent state with skills, sessions, and app-connected tasks. |
| MCP and tools | Built-in MCP management, CLI controls, desktop/dashboard management, health checks, and custom servers. | Extensible tool and plugin connections across the assistant environment. |
| Migration | hermes claw migrate imports supported OpenClaw data with preview and conflict controls. | OpenClaw uses its own onboarding and configuration system. |
| Highlights | Bot Mode, learned skills, CLI/TUI control, MCP management, and provider configuration | Messaging, native apps, paired devices, and cloud-connected personal-assistant tasks |
Key Features
- Bot Mode — Turns profiles into named bots with their own model, memory, skills, credentials, avatar, chat history, routines, and group-chat membership.
- Persistent memory — Keeps reusable information across sessions and exposes memory review through
/journey. - Reusable skills — Turns repeated tasks into skills through the skill system and
/learn. - Scheduled agents — Runs cron jobs and Bot routines;
--continuitypasses the previous output into later runs. - Subagents — Delegates parallel tasks, shows active children, accepts mid-run steering, and can stop a child while preserving partial results.
- Agent-to-agent messaging — Uses Bot chats and
hermes peerto pass work between agents across profiles or registered gateways. - MCP management — Installs and manages MCP servers from the CLI, desktop app, or web dashboard and supports custom HTTP or stdio servers.
- Browser automation — Navigates, clicks, reads, screenshots, and scrolls browser sessions. Authenticated Chromium profiles can be connected with explicit access controls.
- Desktop app — Runs on macOS, Windows, and Linux with chat, sessions, files, profiles, memory, cron jobs, Bots, MCP management, coding projects, and remote connections.
- CLI, TUI, and dashboard — Shares Hermes sessions, configuration, memory, and skills across terminal, desktop, browser administration, and messaging access.
- Messaging gateways — Connects Hermes to Telegram, Discord, Slack, WhatsApp, Signal, email, and other supported channels.
- Model choice — Works with OAuth-backed accounts, API-key providers, custom OpenAI-compatible endpoints, self-hosted inference, and Mixture-of-Agents presets.
- Remote execution — Runs terminal tasks through local, Docker, SSH, Singularity, Modal, Daytona, and other configured backends.
- Coding tools — Includes
/init,/diff,/context,/plan,/goal, worktrees, checkpoints, and project-aware tools. - Grounded research — Supports research with citations and evidence checks when the grounded-citations skill is enabled.
- Voice — Supports streaming TTS, interruption during speech, wake words, and voice messages on supported channels.
- Security controls — Uses command approvals, deny rules, protected instruction-file writes, secret redaction, and sandbox options.
- OpenClaw migration — Imports supported memories, skills, profiles, messaging settings, selected credentials, and workspace instructions through
hermes claw migrate.
Use Cases
- Create specialist bots for research, coding, monitoring, or operations and let them coordinate in Bot Mode.
- Run a recurring news, server, or project monitor that receives its previous report and can avoid duplicate updates.
- Keep project memory across long coding, research, content, or operations tasks.
- Delegate a large task to several subagents and redirect one when its approach needs to change.
- Use Hermes Desktop on a laptop while the agent runtime, provider credentials, and compute run on a remote gateway or server.
- Connect MCP servers for databases, developer tools, observability systems, automation services, and other external capabilities.
- Use browser automation to inspect documentation, collect web information, or complete browser-based tasks.
- Run a personal assistant through Telegram, Discord, Slack, or another messaging platform while the runtime stays on your machine or server.
- Switch providers or self-hosted models for different tasks while preserving existing sessions, skills, and memory.
How to Use Hermes Agent
1. Install Hermes and run setup
On Linux, macOS, WSL2, and Termux, install Hermes through the shell installer.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashNative Windows has a PowerShell installer.
powershell -ExecutionPolicy Bypass -c "irm https://hermes-agent.nousresearch.com/install.ps1 | iex"The Python package option needs the post-install step.
pip install hermes-agent
hermes postinstallReload your shell profile if the installer updated it, then start the setup wizard.
source ~/.bashrc
# Zsh: source ~/.zshrc
hermes setupNous Portal provides OAuth access to Nous services. hermes auth manages credentials and authentication status.
hermes portal
hermes auth status2. Configure a model and open Hermes
Run hermes model from the shell to configure providers, complete OAuth flows, enter API keys, or set a default model.
hermes modelStart the classic CLI, TUI, or desktop app after authentication works.
hermes
hermes --tui
hermes desktop3. Create persistent bots and project context
Bot Mode is built into Hermes Desktop. Each bot is a Hermes profile, and the profile can also be opened from the terminal.
hermes profile create <name>
hermes profile list
hermes -p <bot-name> chatResume an existing session with --continue. In coding repositories, /init can create or update AGENTS.md, /context shows context-window usage, and /diff shows Git changes.
hermes --continue
/init
/context
/diff4. Configure tools, isolation, and MCP servers
Review enabled tools before connecting Hermes to projects, browsers, or messaging channels.
hermes toolsDocker keeps terminal execution in a container. SSH sends terminal work to a configured remote machine.
hermes config set terminal.backend docker
hermes config set terminal.backend sshHermes includes MCP management in the main installation. Open the MCP picker, list available integrations, or install one by name.
hermes mcp
hermes mcp catalog
hermes mcp install <name>5. Connect messaging channels
Configure the messaging gateway after local chat, provider access, and tool permissions are working. The gateway supports channels such as Telegram, Discord, Slack, WhatsApp, Signal, and email. Group and public channels need tighter tool permissions than a private terminal session.
hermes gateway setup
hermes gateway start6. Create scheduled jobs with continuity
Create a recurring job with a schedule and task prompt. --continuity injects the job’s previous output into later runs, which supports deduplication and incremental reporting.
hermes cron create "every 6h" "Scan for new agent-tooling news" --continuity7. Pass work between agents and save repeated tasks
hermes peer registers Hermes gateways on other machines and sends messages to their agents’ canonical Bot Chats.
hermes peer list
hermes peer dm <peer>/<agent> "Review this result and continue the task."For repeated tasks, browse installed skills or turn a completed process into a reusable skill with /learn.
hermes skills
/learn <process>8. Update, export, or migrate an existing setup
Use the built-in updater for an existing installation. Session exports can include a secret-redaction pass before you share them.
hermes update
hermes sessions export backup.jsonl --redacthermes claw migrate imports supported OpenClaw data. Run a dry preview before writing migration changes.
hermes claw migrate --dry-runHermes Agent Cheatsheet & Command Reference
Installation Commands
| Command | Purpose |
|---|---|
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash | Install on Linux, macOS, WSL2, or Termux. |
powershell -ExecutionPolicy Bypass -c "irm https://hermes-agent.nousresearch.com/install.ps1 | iex" | Install on native Windows through PowerShell. |
pip install hermes-agent | Install the Python package. |
hermes postinstall | Run post-install setup after the Python package installation. |
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash -s -- --skip-browser | Install while skipping browser automation setup. |
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash -s -- --include-desktop | Install Hermes with desktop support. |
hermes desktop | Launch Hermes Desktop. |
hermes update | Update an existing Hermes installation. |
Core CLI Commands
| Command | Purpose |
|---|---|
hermes | Start the classic interactive CLI. |
hermes --tui | Start the terminal UI. |
hermes desktop | Open the native desktop app. |
hermes model | Configure providers, authenticate, set endpoints, and choose the saved default model. |
hermes auth | Manage credentials and OAuth-backed provider authentication. |
hermes tools | Configure enabled tools. |
hermes config | View or change Hermes configuration. |
hermes gateway setup | Configure messaging platforms. |
hermes gateway start | Start the configured gateway. |
hermes setup | Run the setup wizard. |
hermes portal | Run the Nous Portal setup flow. |
hermes status | Show agent, authentication, and platform status. |
hermes doctor | Diagnose setup problems. |
hermes --continue / hermes -c | Resume the most recent session. |
hermes skills | Browse and manage reusable skills. |
hermes mcp | Open the MCP picker and manage MCP server configurations. |
hermes cron | List, create, edit, pause, resume, run, or remove scheduled jobs. |
hermes peer | Manage peer gateways and direct agent-to-agent messaging. |
hermes claw migrate | Import supported OpenClaw data into Hermes. |
hermes sessions export backup.jsonl --redact | Export session data with a secret-redaction pass. |
Bot, Cron, and MCP Commands
| Command | Purpose |
|---|---|
hermes -p <bot> chat | Open the persistent profile used by a Bot Mode bot. |
hermes profile create <name> | Create a Hermes profile that can also appear as a bot. |
hermes profile list | List profiles. |
hermes peer list | List registered peer gateways. |
hermes peer dm <peer>/<agent> "..." | Send a direct message to an agent on a registered peer. |
hermes cron list | List scheduled jobs and Bot routines. |
hermes cron create "every 2h" "..." | Create a recurring scheduled job. |
hermes cron create "every 6h" "..." --continuity | Create a recurring job that receives its previous output on later runs. |
hermes cron edit <job_id> --continuity | Enable continuity on an existing job. |
hermes mcp | Open the interactive MCP picker. |
hermes mcp catalog | List available MCP entries. |
hermes mcp install <name> | Install an MCP server by name. |
hermes mcp add <name> ... | Configure a custom HTTP or stdio MCP server. |
hermes mcp serve | Run Hermes itself as an MCP server. |
Slash Commands
| Command | Purpose |
|---|---|
/new | Start a new conversation. |
/model | Switch between providers and models that are already configured. |
/status | Show session, model, provider, token, and recent activity information. |
/context [all] | Show context-window usage by system prompt, tools, skills, MCP, subagents, memory, conversation, and free space. |
/diff [staged|all|session] [--stat] | Show Git changes for the current project. |
/init [notes] | Generate or update AGENTS.md from a repository scan. |
/plan [task] | Write a Markdown implementation plan without executing it. |
/focus [on|off|status] | Show a reduced-output view with the prompt and final response while hiding routine tool progress. |
/learn <task> | Create a reusable skill from a task description or source. |
/journey | Review saved memories and learned skills. |
/goal <text> | Set a persistent completion goal that Hermes can continue working toward across turns. |
/agents | Show active agents and running tasks. |
/cron | Manage scheduled tasks from chat. |
/browser | Connect, disconnect, or inspect a local Chromium-family browser connection. |
/compress | Compress conversation context. |
/usage | Show token usage, estimated cost, and available provider-account limits when supported. |
/rollback | List or restore filesystem checkpoints. |
/voice on | Enable CLI voice mode when speech dependencies are configured. |
/undo | Remove the last exchange. |
Configuration Examples
| Command | Purpose |
|---|---|
hermes config set terminal.backend docker | Use Docker isolation for terminal execution. |
hermes config set terminal.backend ssh | Use SSH for remote terminal execution. |
hermes config set cron.model <model> | Set a default model for unpinned cron jobs. |
Provider and Model Access
Hermes supports OAuth-backed accounts, API-key providers, model aggregation services, custom OpenAI-compatible endpoints, and self-hosted models. Use hermes model for the current provider setup flow.
| Provider or Setup | Access Method |
|---|---|
| Nous Portal | OAuth through hermes model or hermes portal. |
| OpenAI Codex | ChatGPT/Codex OAuth through hermes model. |
| Anthropic | Supported OAuth flows or an Anthropic API key. |
| GitHub Copilot | OAuth/device authentication through hermes model. |
| OpenRouter | OPENROUTER_API_KEY. |
| Google Gemini / Vertex AI | Provider credentials through the relevant Gemini or Google Cloud setup. |
| z.ai / GLM | GLM_API_KEY or supported z.ai credentials. |
| Kimi / Moonshot | KIMI_API_KEY or the China-region credential for that provider. |
| MiniMax | API-key or supported OAuth setup. |
| xAI Grok | API-key or supported OAuth setup. |
| Alibaba Cloud Qwen | DashScope or Alibaba coding-plan credentials. |
| Hugging Face | HF_TOKEN. |
| Azure Foundry | Azure Foundry endpoint credentials. |
| Tencent TokenHub / TokenPlan | Tencent provider credentials through the supported provider configuration. |
| CommandCode | Configure through hermes model. |
| Nebius Token Factory | NEBIUS_API_KEY. |
| Ramp Router | RAMP_ROUTER_API_KEY. |
| Custom / self-hosted | Any compatible OpenAI-style endpoint, including local or self-hosted inference servers. |
Alternatives & Related Resources
- OpenClaw 2.0: Personal AI Assistant for Apps, Devices & Cloud Work
- 7 Best OpenClaw Alternatives for Safe & Local AI Agents
- 10 Best Agent Skills for Claude Code & AI Workflows
- Most Popular Agent Skills on GitHub for Coding Agents
- Hermes Agent Documentation
- Migrate from OpenClaw
Pros
- Free and open-source under the MIT license.
- Persistent memory and reusable skills.
- Bot Mode for persistent specialist agents.
- Cron continuity for recurring jobs.
- Parallel subagents with mid-run steering.
- Native desktop app for macOS, Windows, and Linux.
- CLI, TUI, dashboard, messaging, and remote gateway access.
- Multiple model providers and custom endpoints.
- Built-in MCP management and browser automation.
- OpenClaw migration tools.
Cons
- Tool and browser permissions need careful review.
- Multi-provider and multi-agent tasks can increase model costs.
- Remote gateways require careful credential handling.
- Long-running automations need maintenance and monitoring.
FAQs
Q: Is Hermes Agent free?
A: Hermes Agent is free and open-source under the MIT license. Model APIs, hosted inference, cloud machines, browser services, and other third-party services can have their own costs.
Q: Does Bot Mode require another Hermes installation?
A: No. A Bot Mode bot is a Hermes profile with its own configuration, memory, skills, credentials, and chat history. The desktop app presents the profile as a bot, and hermes -p <bot> chat can open that profile from the terminal.
Q: Do Hermes cron jobs remember previous runs?
A: A job created or edited with --continuity receives its previous output on later runs. Cron agents can also use persistent memory, which supports recurring monitors and incremental reports.
Q: Can Hermes use a browser profile where I am already signed in?
A: Hermes can connect to a Chromium-family profile for browser automation. That connection can expose signed-in sessions to the agent, so enable it only for accounts and sites you are prepared to let Hermes access.
Q: Is Hermes Agent private?
A: Hermes can keep configuration, memory, skills, and local tasks on your machine or server. Data leaves that environment when a task uses external model providers, cloud execution, browser services, web tools, MCP servers, or remote gateways that transmit it elsewhere.
Recent Hermes Agent Releases
- August 31, 2026 · v0.21.0. Bot Mode, peer agent messaging, cron memory and continuity, steerable subagents, MCP management, desktop browser control, new providers, and security hardening.
- August 3, 2026 · v0.20.0. Conversational voice, wake words, grounded citations, signed webhooks, desktop artifacts, plugin SDK, A2A support, coding commands, and improved tool recovery.
- July 20, 2026 · v0.19.0. Faster first turns, live reasoning streams, desktop performance changes, secret sources, smart approvals, live subagent transcripts, durable background delegation, and session export.
