Compare · Cloud agent · Checked November 12, 2026

Hermes Agent vs Instinct

You change how Hermes Agent behaves, from its persona to how often it checks in, while Instinct ships one behaviour its vendor sets. Nous Research's free, open-source agent answers on more than 20 messaging platforms. Instinct's terms, revised Aug 26, 2026, forbid modifying the service.

Researched by the Nous Research team · Updated November 12, 2026

Key differencesHermes AgentInstinct
Getting in
Install it today
Invite-only
Where it runs
Your computer or the cloud
Instinct's cloud
Models
You pick
No picker in its docs
Behaviour
Yours to change
As the vendor ships it

Why Hermes Agent

  1. [ 01 ]

    No invite and no waitlist

    Install Hermes Agent today under the MIT license. Instinct is invite-only: its sign-in terms describe a waitlist and a welcome text when a spot opens.

  2. [ 02 ]

    It runs on your own computer

    Hermes Agent runs on your computer, a server or Hermes Cloud, and stores memory and skills on that machine. Instinct runs on its own cloud computer.

  3. [ 03 ]

    You choose the model

    You pick the provider and model for each request, and switch mid-session. Instinct documents no model picker and names no model it runs.

  4. [ 04 ]

    You change how it behaves

    Change the agent's persona, its reminder cadence and what it does on every event. Instinct's terms forbid modifying the service or making derivative works.

Which to pick

Instinct
Instinct fits if you have an invite and want its behaviour as shipped, with the vendor choosing the model and setting how the agent acts.
Hermes Agent
Choose Hermes Agent when you want to shape the assistant: its persona, its cadence, its tools and the model behind it.

At a glance

A check marks where Hermes Agent has the edge.
Hermes Agent and Instinct compared by topic
TopicHermes AgentInstinct
Source codeHermes Agent:Edge: Open source under the MIT license, with its code on GitHub.Instinct:Proprietary, with no source code or open-source license published.
Where it runsHermes Agent:Edge: Your computer, a server or Hermes Cloud.Instinct:Instinct's own cloud computer.
ModelsHermes Agent:Edge: Frontier models on launch day, or a local model you run.Instinct:No model picker in its docs, and no model named.
Chat appsHermes Agent:Edge: iMessage, WhatsApp, Telegram, Slack, Signal, SMS, email and more.Instinct:iMessage, WhatsApp, SMS and calls.
Your dataHermes Agent:Edge: Stored on the machine that runs it. No telemetry by default.Instinct:Trains on what you send unless you opt out. Google and Vault data excluded.
Where you can use itHermes Agent:Edge: A public download, with no invite and no waitlist.Instinct:Invite-only, with a waitlist when it is at capacity.
Tools and extensionsHermes Agent:Edge: Persona, cadence, hooks, plugins and tools are all yours to edit.Instinct:Its terms forbid modifying the service or making derivative works.

Switch from Instinct

  1. Install Hermes Agent

    Get Hermes Desktop for macOS (Apple Silicon), Windows and Linux, or start an always-on agent on Hermes Cloud.

  2. Pick the model it runs on

    Run hermes setup --portal to sign in and pick a model, or point it at a local model through Ollama or vLLM.

  3. Connect your messages and Google

    Run hermes gateway setup for iMessage, WhatsApp or Telegram, and link a Google account to act on Gmail, Calendar and Drive.

hermes setup --portal

In detail

You can install Hermes Agent today

Hermes Agent is a public download. Install Hermes Desktop on macOS, Windows or Linux, or the CLI with one command, and start an always-on agent on Hermes Cloud from any paid Nous Portal plan. There is no invite, no waitlist and no approval step.

Instinct is invite-only. Its SMS terms, last revised on Aug 17, 2026, describe signing up while Instinct is at capacity and getting a welcome text once a spot opens, and say the sign-in consent covers waitlist updates. The Verge reported on Oct 9, 2026 that access needs an invitation from a current user or waitlist approval.

Your computer, a server or Hermes Cloud

Hermes Agent runs where you install it: your own computer, a server or a VPS. Its shell commands can run locally, in Docker, over SSH, or in Modal, Daytona or Vercel Sandbox. Hermes Cloud is the always-on option, running 24/7 in its own container and opening in a browser tab with the Hermes Desktop interface, and every paid Nous Portal plan includes one.

Instinct runs on its own cloud computer. Its privacy policy, last revised on Aug 26, 2026, describes an autonomous personal assistant with Mac OS and mobile apps, and its home page says it can use a phone and computer the way you do. Spear Street Technology documents no way to run it on hardware you control.

Hermes Agent stores memory, skills and sessions on the machine that runs it, and model requests go to the provider you pick.

You pick the model, not the vendor

Hermes Agent sends each request to the provider and model you configure. One Nous Portal login covers hundreds of models, including Claude, GPT and Gemini, with frontier models on launch day. It also works with OpenRouter, any OpenAI-compatible API and local models through Ollama or vLLM.

The /model command switches models mid-session, and conversation history, memory and skills carry over. In Hermes Desktop, Bot Mode gives each named Bot its own model, so a Claude Bot and a GPT Bot can share a Group Chat.

Instinct names no model and documents no model picker. Its privacy policy says it shares information with third-party AI model providers to operate the assistant, so the choice is Instinct's, not yours.

Both answer in your messages, on more platforms here

Both products answer where you already type. One Hermes Agent gateway connects more than 20 messaging platforms, including iMessage, WhatsApp, Telegram, Discord, Slack, Signal, Microsoft Teams, SMS and email. The gateway owns every conversation, so a chat you open in the CLI, Hermes Desktop or a messaging app is the same chat.

Instinct works over iMessage, WhatsApp, SMS and calls, and Noah Shinn said on Oct 27, 2026 that group chats reached everyone on iMessage. Hermes Agent Group Chats put 2 to 6 named Bots in one thread, each on its own model, and its scheduled jobs are written in plain language and report to any chat.

Your data stays on the machine you run

A personal agent needs your accounts, so where the data sits matters. Hermes Agent keeps memory, skills and sessions on the machine that runs it, collects no telemetry by default, and holds credentials in your own keychain or env file. You read the code that uses them.

Instinct's privacy policy says it uses your information to evaluate, fine-tune and train the AI models behind its products, with an opt-out in settings that it says applies going forward and not to other improvements. It excludes what it receives from Google Workspace, and its terms exclude material you place in its Vault. It also says disconnecting an integration does not by itself delete the data already collected from it.

Instinct's terms, last revised on Aug 26, 2026, appoint the service as your agent to enter agreements and transactions on your behalf. Hermes Agent asks before it acts outside your machine, and its approval gate is in the code.

You change how it behaves, down to its cadence

A personal assistant you cannot adjust is a personal assistant on someone else's terms. SOUL.md sets who your Hermes Agent is and is the first thing in its system prompt, so you rewrite its character in a text file. /personality switches a preset for one session. If its check-in rhythm annoys you, /heartbeat sets the interval for a recurring prompt, and cron jobs written in plain language run on the schedule you name.

Event hooks run your own code at lifecycle points, so you can block a tool call, rewrite content or push events to your own services. Plugins add tools and integrations without touching core code, and the Hermes Marketplace has more than 1,000 of them. All of it is MIT-licensed code on GitHub, so nothing stops you from changing the parts no setting exposes.

Instinct ships one behaviour. Its terms, last revised on Aug 26, 2026, forbid modifying the service or creating derivative works from it, and the only settings its own pages mention are an account page and a training opt-out. Spear Street Technology publishes no source code, no plugin system and no API.

The Internet's Own AI

© 2026, Nous Research, Inc.

Terms|Privacy

MIT License · 2026