Enjoy waived service fees on top-ups, plus exclusive referral gifts
Open SourceMIT License

Hermes Agent

An agent that grows with you.

Hermes Agent is an open-source long-lived agent with persistent memory, a learning loop, scheduled automations, isolated subagents, and multiple execution backends.

Surfaces & capabilities

CLITelegramDiscordSlackWhatsAppSignalEmailBrowserDockerSSHMCPSubagents

Capabilities

From chat session to operating agent

Persistent Memory

Hermes remembers prior sessions, stores working context, and builds a more useful operating model over time instead of starting from zero every run.

Learning Loop

It can turn repeated workflows into reusable skills, improve how it works, and reduce repetition across future sessions.

Scheduled Automations

Run recurring tasks and natural-language schedules so the agent can keep working even when you are not actively prompting it.

Parallel Subagents

Delegate research and execution into isolated subagents so large tasks can be split without bloating the main conversation context.

Sandboxed Execution

Use local, containerized, remote, or isolated execution backends for safer tool use, system access, and controlled automation.

Multi-Surface Access

Operate Hermes from terminal, messaging channels, gateways, and connected tools while keeping one long-lived agent runtime behind the scenes.

Better together

Run Hermes on one governed endpoint across more models

Hermes supports configurable providers and custom endpoints. By pointing it to OneRouter, one workflow can access Claude, GPT, Gemini, DeepSeek, MiniMax, and more with governance, billing, and switching policy layered on top.

One endpoint across production models
Unified governance and team controls
Safer provider switching over time
Get a OneRouter API Key

Install Hermes

bash
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Setup and model selection

bash
hermes doctor
hermes setup
hermes model

Example: point Hermes at a unified model endpoint

bash
# guided setup
# choose your provider or custom endpoint in hermes model
# then launch the agent
hermes

If you use the MiniMax Token Plan, you can select MiniMax directly in `hermes model`. If you want broader multi-model access, switch the provider flow to a OneRouter-backed endpoint strategy.

Typical workflow

From install to continuous operation

A common Hermes flow is to install, validate the environment, complete setup, choose a model, enable tools and gateways, then turn repeated work into skills and scheduled automations.

1

1. Install and run `hermes doctor`

2

2. Complete `hermes setup` and pick a provider

3

3. Use `hermes model` to set your default model

4

4. Launch `hermes` and begin working

5

5. Add memory, skills, tools, and schedules over time

Core commands

bash
hermes
hermes setup
hermes doctor
hermes model
hermes tools
hermes gateway
hermes update

These commands cover the main Hermes lifecycle: post-install validation, setup, model switching, tool enablement, gateway usage, and updates.

FAQ

Frequently Asked Questions

Hermes is designed as a long-lived agent runtime with memory, automation, and delegation. It is closer to an operational agent than a single-session chat tool.
Yes. Hermes supports configurable model providers and custom endpoints. Pointing it to OneRouter lets one setup access a broad set of production models through a unified endpoint.
No. The CLI is the main surface, but Hermes also supports gateway and messaging-oriented workflows, plus tool and environment integrations.
OneRouter simplifies model access, governance, and switching. That makes Hermes easier to operate across teams without hard-coding one provider into every workflow.

Put Hermes into your production workflow

Connect a long-lived agent to OneRouter so you can run more automation within a unified model governance and team operations layer.