What Is Hermes Agent? The Free AI Agent That Remembers Everything

What Is Hermes Agent? The Free AI Agent That Remembers Everything

Imagine an assistant that does not forget you every time you close the window. You tell it your name, your project and your preferences on Monday, and on Friday it still remembers — it picks up where you left off, finishes the job you started. It lives on your own computer, answers you on Telegram, Discord, Slack or email, and works with almost any AI model you choose, including free models on your own machine.

That is Hermes Agent: a free, open-source AI agent from Nous Research that runs on your laptop or a cheap server, remembers what it learns across sessions, and reaches you through 25+ messaging platforms. Released in February 2026, it is one of the fastest-growing open-source projects of the year: as of 15 November 2026, NousResearch/hermes-agent has 263,515 stars and 59,401 forks, with latest stable release v0.23.1 published on 10 November 2026.


What Is Hermes Agent, in Plain Words?

A chatbot answers one question at a time and then forgets everything. An agent can plan a multi-step job, use tools, check its own work, and keep going until the job is done. Hermes Agent is that kind of program — it runs on your own machine, not inside a company's app, and you can talk to it from almost anywhere.

Think of it as a personal assistant with a very good memory. Install it once on your computer or a small server, connect an AI model — Claude, GPT, GLM, DeepSeek, or a free local model — and talk to it as you talk to people: a Telegram message, a Discord chat, an email, a terminal command. It reads your message, decides what to do, uses its tools, and reports back. If a job takes hours, it keeps working in the background and messages you when it is finished.

The part that makes it special is memory. Most AI assistants start every conversation from zero. Hermes Agent keeps notes about you, your projects and your preferences, and remembers them across sessions — days and weeks later. The longer you use it, the more useful it becomes. It costs nothing to install: Hermes Agent is free and open source under the MIT licence. You only pay for the AI model it calls — and a free local model costs nothing at all.

Why Is Everyone Talking About Hermes Agent?

The short answer is the numbers. As of 15 November 2026, Hermes Agent has 263,515 stars — more than OpenCode, covered in our 14 November post — and it passed 100,000 stars within about ten weeks of its February 2026 release, per Decrypt. A project that grows this fast, with this many people watching it, is not a hobby experiment — it is infrastructure that thousands of people are building on.

The second reason is timing. In 2026, personal AI assistants split into two camps: closed assistants inside a company's app, and open agents you run yourself. Hermes Agent is the most popular open agent of the second. It gives you the memory and the always-on behaviour of a paid assistant, but the software is free, the data stays on your machine, and you choose the model.

The third reason is what it can reach. Hermes Agent talks to you through 25+ messaging platforms and works with 20+ model providers. Most tools lock you into one model or one chat app; Hermes Agent sits in the middle, connects to almost anything, and lets you swap the model whenever you like. That flexibility is why it is the default agent harness for people who run AI on their own hardware, including Ollama users (see our 13 November post).

What Can Hermes Agent Actually Do? Its Functions Explained

Here is what each function means in plain language.

1. It Remembers You Across Sessions

Hermes Agent keeps a persistent memory of your conversations, preferences, projects and facts you tell it — ask it to remember a short-summary preference or a 1 December deadline, and it will still know next week. The memory is stored on your own machine, not on a company's server, so you control it and can read or delete it at any time.

2. It Writes Its Own Skills

When Hermes Agent works out a recurring job, it saves the steps as a reusable skill and reuses it next time instead of starting from scratch. Version 0.23.0 trimmed built-in skills from 58 to 44 (Word, PowerPoint and Excel merged into one microsoft-office skill).

3. It Reaches You on 25+ Messaging Platforms

It can answer you on Telegram, Discord, Slack, WhatsApp, email and 20+ other platforms, plus a command line and a desktop app. Since v0.23.0, "The Hearth Release" of 5 November 2026, one gateway owns every session, so a chat started in the terminal can continue on Telegram.

4. It Works with 20+ AI Models and Providers

Hermes Agent is model-agnostic: Claude, GPT, GLM, MiniMax, DeepSeek, Gemini and others, plus a local model through Ollama, vLLM or LM Studio. Since v0.23.0, GLM-5.3 is the new default model.

5. It Runs Jobs on a Schedule

Hermes Agent runs tasks on its own, on a timetable you set. Describe the job once — a daily news summary, a weekly report — and it repeats it, sending the result to your chat app. This is the difference between a chatbot and an agent: a chatbot waits for you, while Hermes Agent can work while you are asleep.

6. A Desktop App — Now in Your Browser Too

As well as the command line, Hermes Agent has a desktop app for Mac and Windows. Since v0.23.0, that workspace can also open in a browser tab served by your own gateway — off by default, and needing macOS 13+ or 64-bit Windows.

7. A Plugin Catalog of 1,050+ Entries

Community plugins and MCP tools extend Hermes Agent — web search, calendars, specialised work tools. In v0.23.0 the plugin catalog grew from around 600 entries to more than 1,050, with a Verified tier for trusted entries. Paid MCP tools can declare their price tag, and you can set spend limits so a tool cannot quietly run up a bill. Think of it as an app store for your agent, with the spending controls switched on.

8. It Updates Itself and Keeps a Backup

Hermes Agent updates itself and backs up your configuration first, so an update should not wipe your settings. That matters because the project moves fast — ten releases landed between 12 October and 5 November 2026 alone. The desktop app installed from the Microsoft Store updates through the Store, and the Docker image updates when you pull the new stable tag.

FunctionWhat it doesExample in daily use
Persistent memoryRemembers you, your projects and your preferences across sessionsDeadline known next week
Self-written skillsSaves the steps of a recurring job as a reusable skillSame weekly report every Monday
25+ messaging platformsTelegram, Discord, Slack, WhatsApp, email, moreJob started on laptop, checked on phone
20+ model providersClaude, GPT, GLM, DeepSeek, local OllamaCheap model routine, strong model hard jobs
Scheduled jobsRuns tasks on a timetable you set, and messages you the resultDaily news summary at 8 AM
Desktop + browser workspaceChat, terminal, memory in an app or browser tabCheck your agent from any device
Plugin catalog (1,050+)Community tools, MCP servers, spend limitsWeb search or calendar, no coding
Self-updates with backupUpdates itself and backs up your configuration firstStay current safely

How Hermes Agent Works, Step by Step

Follow one request through the system: you message Hermes Agent on Telegram, "Summarise my unread emails and tell me which ones need a reply today." First, your message reaches the gateway, the single hub that v0.23.0 introduced. The gateway owns every session, so whether you wrote from Telegram, the terminal or the desktop app, the conversation lands in the same place, with the same memory attached.

Second, the gateway hands your request, with the relevant memories and skills, to your chosen model, which plans the job. Third, the agent carries out the plan with its tools. Fourth, it replies on Telegram and saves what it learned — that you want email summaries every morning. The next morning, it can do the same job on schedule without being asked. Your data stays on your machine throughout: the gateway, memory and tools all run on hardware you control.

Memory and Skills: The Part That Makes It Different

Most AI tools now claim memory, so here is what Hermes Agent's actually is. It is not one big chat log. The agent keeps structured notes — facts about you, decisions you made, preferences you stated, lessons from completed jobs — and pulls the relevant notes into each new conversation. That is why it can remember your deadline three weeks later without re-reading every message you ever sent.

A skill is a saved recipe: the steps, tools and checks for a job the agent has done before. Hermes Agent writes these itself as it works, and you can also add skills made by other people from the plugin catalog. The combination is what makes the agent compound: in week one it is a clever chatbot; by week four it knows how you like your reports and what "urgent" means to you.

Memory is only as good as what you put into it, and it needs occasional tidying: if you change a preference, say so clearly, or the old note can linger. Back it up like any other important file — it lives on your machine.

Providers and Models: What You Can Plug In

Hermes Agent does not come with a brain built in — you bring your own, from 20+ providers. The table covers the routes most people use.

RouteWhat it isBest for
Nous PortalOne sign-in, hundreds of models, hosted toolsSimplest start, no per-provider keys
Anthropic ClaudeClaude API keyLong, careful work and writing
OpenAI GPTGPT API keyGeneral work, strong tool use
z.AI GLMGLM API key; GLM-5.3 is the new defaultGood value, strong all-round performance
DeepSeekDeepSeek API keyLow-cost coding and reasoning
MiniMaxMiniMax API keyLow-cost general use
Local Ollama / vLLM / LM StudioRuns on your own computer, freePrivate work, zero model bill

For readers in Pakistan, the local route is worth a note: if you already run Ollama — see our 13 November post for the setup — you can point Hermes Agent at that local model and pay nothing for the AI. The trade-off is speed and quality, so many people mix the two: local for private, routine jobs, paid for the hard ones. Switching is a settings change, not a reinstall.

How to Install Hermes Agent

On macOS or Linux, install with a single terminal command from the project's own install address (hermes-agent.nousresearch.com/install.sh). Windows users can use the PowerShell installer or the Microsoft Store, and there is a Docker image — nousresearch/hermes-agent:stable — for servers. Then run hermes version: it should print v0.23.1 or newer.


Then choose a model. The simplest path is a free Nous Portal account: sign in once, and the agent can use the models and hosted tools without a separate key per provider. Or bring your own API key from Anthropic, OpenAI, z.AI, MiniMax or DeepSeek, or point the agent at a local model through Ollama. Either way the software itself costs nothing: it is free under the MIT licence, and your only bill is your model provider's charge.

A first setup takes about fifteen minutes: install, sign in or add a key, pick a model, connect one chat app (Telegram is easiest) and give the agent one small job, such as a daily summary. Get that working before adding more platforms, plugins or a server. On a server, use a cheap VPS, keep it updated, and never expose the gateway to the public internet without a password.

What Does Hermes Agent Cost?

The software is free; the model is what costs money, and you choose how much to buy. Prices below are as listed on the Nous Portal on 12 November 2026 — check before paying, because prices and limits change.

OptionPriceWhat you get
Hermes Agent softwareFree (MIT licence)Full agent: memory, skills, messaging, scheduling
Local model (Ollama / vLLM)FreeRuns on your computer; no model bill
Nous Portal Free$0Free models only, standard limits, no credits
Nous Portal Plus$20 a monthPaid models, hosted tools, one Cloud agent (4 GB/25 GB), $22 credits
Nous Portal Super$100 a monthFive times the Plus usage, one Cloud agent (4 GB/100 GB), $110 credits
Nous Portal Ultra$200 a monthTen times the Plus usage, one Cloud agent (8 GB/100 GB), $220 credits
Your own API keyPay per useClaude, GPT, GLM, DeepSeek or others, billed by the provider

Two details matter. First, each paid plan includes slightly more in usage credits than its price — Plus includes $22 for $20 — with a rollover cap ($10 Plus, $50 Super, $100 Ultra). Second, included usage is not unlimited: the Portal applies daily, weekly and monthly limits it does not publish as numbers, and pricier models consume the allowance faster. The Cloud agent in every paid plan is a hosted, always-on agent for people who do not want to run their own machine around the clock. Fees are non-refundable, so start with the free tier or a small top-up before committing.

The cheapest serious setup is a local model through Ollama: no monthly bill, just your own computer and electricity. The most convenient is Nous Portal Plus at $20 a month. Choose by whether you value zero cost or zero setup more.

Hermes Agent vs OpenClaw vs OpenCode

Three open-source AI projects dominate this space, and they are easy to confuse: all three are free, model-agnostic and hugely starred — but built for different jobs.

Hermes AgentOpenClawOpenCode
What it isSelf-hosted agent with persistent memoryPersonal AI assistantAI coding agent
Stars (Nov 2026)~263,000~394,000~225,000
Main strengthMemory, skills, messaging, schedulesAlways-on personal helpWriting and editing code
Talk to it on25+ platformsMessaging appsTerminal and editor
Local modelsYes — Ollama, vLLM, LM StudioYes — local models supportedYes — local Ollama backend
Best forDaily assistant workPersonal productivitySoftware development

Star counts are dated 15 November 2026 for Hermes Agent (263,515, checked live), early November for the other two. The practical difference is simpler: OpenClaw is a personal assistant, Hermes Agent is the agent harness — the system that runs jobs, remembers things and reaches you anywhere — and OpenCode is a coding specialist. If your main job is writing code, start with OpenCode; for everything else, start with Hermes Agent.

Who Should Use Hermes Agent — and Who Should Not

Hermes Agent fits if you want an assistant that remembers you and works while you are away; if you already run local AI through Ollama and want an agent on top of it; if you are comfortable with a terminal and a settings file; or if you want your data on your own machine, not a company's server. It also suits small businesses wanting scheduled summaries and inbox triage.

It is not a good fit if you want zero setup — a closed assistant app is simpler, and Hermes Agent still needs installing, a model key and occasional maintenance. Nor if you need guaranteed stability: it ships fast and sometimes breaks things, so it suits people who can tolerate an occasional hiccup, not a business process that must never fail.

Honest Limits and Risks

No honest review can skip the pace of change: ten releases landed between 12 October and 5 November 2026, and that speed sometimes breaks things. Version 0.23.0 removed the old browser tools in favour of browser_exec, renamed hermes mcp serve to hermes mcp-serve, and changed the built-in skills — any of which can break a working setup. If you depend on the agent daily, pin a stable release and read the release notes before updating.

Security is your responsibility, because you are the host. The public advisory database lists ten 2026 advisories, all in versions up to 0.19.0, the highest rated CVSS 7.5. Version 0.23.1 is past those — but only if you actually update. Keep it updated, keep its gateway behind a password, and never expose it to the public internet without authentication. Daily canary builds are for testers; everyone else should stay on stable releases.

There are cost risks too: an agent that runs jobs on its own can spend money on its own — a looping scheduled job, a paid plugin, a powerful model left as default. That is why v0.23.0 added price tags for paid MCP tools and spend limits: use them. Set a spending cap before enabling scheduled jobs, and check usage in the first week.

Finally, Hermes Agent is a harness, not a brain: if your model makes a mistake, the agent will confidently carry it out. It can browse, message people and change files for you, so start with narrow permissions — read-only access, no sending messages without approval — and widen them only once you trust it. One claim to be careful with: some coverage mentions "forgetting 2.4x less" — that belongs to Meta's research, not to Nous, and this post does not repeat it as a Nous claim.

Frequently Asked Questions

1. Is Hermes Agent really free?

Yes. The software is free and open source under the MIT licence. What costs money is the AI model it calls — a local model through Ollama is free too. A Nous Portal plan or a provider API key is optional; you only pay for what you use.

2. What is the latest version of Hermes Agent?

As of 15 November 2026, the latest stable release is v0.23.1 (10 November 2026); the previous major release, v0.23.0 "The Hearth Release", came out on 5 November 2026. Check the current version at publish time — this project releases very often.

3. Can Hermes Agent run on my own computer with a free local model?

Yes — through Ollama, vLLM or LM Studio, the model runs on your own machine and costs nothing. A small model runs on an ordinary laptop; larger ones need more memory and a graphics card. Our 13 November post explains the Ollama setup.

4. How is Hermes Agent different from OpenClaw?

OpenClaw is a personal assistant; Hermes Agent is the agent harness. Both are free, open source and model-agnostic, and both can use local models. Hermes Agent's emphasis is persistent memory, self-written skills, scheduled jobs and 25+ platforms through one gateway — the stronger fit for an agent that runs jobs around the clock and remembers everything.

5. Which messaging apps can I use to talk to Hermes Agent?

More than 25, including Telegram, Discord, Slack, WhatsApp and email, plus a command line and desktop app. Since v0.23.0, one gateway owns every session, so a conversation can move between platforms without losing history. Start with Telegram and add more once it works.

6. Do I need Nous Portal to use Hermes Agent?

No. Nous Portal is Nous Research's account and billing layer: one sign-in for hundreds of models, hosted tools and a Cloud agent. It is the simplest way to start and has a free tier, but it is optional — you can use your own API key or a free local model instead.

7. Is it safe to let an AI agent run on my computer?

As safe as the care you put into it. It runs on your machine, your memory stays there, and the code is open source. But you are the administrator: keep it updated — ten 2026 advisories exist, all in versions up to 0.19.0, highest CVSS 7.5 — protect the gateway with a password, start with narrow permissions, and set a spending limit first.

The Bottom Line

Hermes Agent is the most complete answer yet to a simple question: what if your AI assistant remembered you, lived on your own computer, and worked while you were away? At 263,515 stars as of 15 November 2026, free under the MIT licence, with 25+ platforms and 20+ providers, it is the most-starred agent harness in open source — ahead of OpenCode, behind only OpenClaw.

It is not a finished product: it ships fast, sometimes breaks things, and asks you to be your own administrator. But install it, pick a model (a free local one through Ollama is the cheapest way in) and give it one small recurring job, and within a week you will see the point: an assistant that gets more useful the longer you use it, because it remembers. That is the whole idea, and it is free.

By Umar Farooq, Tech Know.

Post a Comment

0 Comments