Stats
We may earn commissions from some links.We may earn commissions from links to support our work. Learn more.
Get more AI tool alerts:
About Hermes Agent
Hermes Agent is Nous Research's free, open-source AI agent for getting work done with tools, persistent memory and reusable skills. It can work with files, run commands, research the web and handle scheduled tasks. You can use it through a terminal, the official desktop app or connected messaging services.
I'd shortlist it if you keep teaching AI the same working procedure. A useful research process, reporting format or deployment method can become a saved skill rather than another long prompt you paste into every conversation.
The software is free. Models, paid tools and optional hosting are separate expenses. That flexibility is attractive, but you are choosing how to operate an agent, not buying an unlimited chat subscription.
Explore Hermes Agent and its setup options.
What is Hermes Agent?
Hermes is the software around a language model that lets it take action. The model decides what to do; Hermes supplies tools, manages the conversation and provides ways to retain useful knowledge between sessions.
It is separate from Nous Research's Hermes model family. You can connect other providers or a compatible local model. Choosing Hermes Agent does not commit you to using a Nous model.
For example, you could ask it to research suppliers, save a comparison document and turn the successful procedure into a skill for the next search. That combines tool execution with reusable instructions. Whether the report is good still depends on the model, sources and instructions; saving a procedure does not validate its contents.
Who is Hermes Agent For?
Hermes makes the most sense for developers, researchers and technical operators with recurring work. It is especially relevant when the work crosses applications: gathering information, editing files, using a browser and delivering the result somewhere else.
A creator might want a repeatable research brief. A developer might want project conventions retained between tasks. An operator might want a scheduled check delivered to Telegram instead of another dashboard to visit. These are sensible uses of its documented tools, not reasons to assume every workflow runs unattended without adjustment.
If your goal is simply to build a website or business app, our guide to AI app builders and vibe coding tools covers a more focused category. Hermes is a general agent you configure around your work, rather than a dedicated app-building service.
I would skip the setup for occasional questions or short rewrites. The strongest reason to adopt Hermes is having procedures worth keeping and running again.
Hermes Agent Pros and Cons
The trade-off is control versus operating effort. Portal and Cloud can simplify parts of the setup, but they do not remove the need to define a useful task and inspect its result.
Hermes Agent Features: Memory, Reusable Skills and Specialist Bots
Skills that preserve how you work
Skills are Markdown instructions the agent loads when relevant. They can describe a research method, a technical procedure or a preferred output format, with supporting material where needed. Hermes supports bundled, installed and agent-created skills.
The /learn command gives this feature a practical entry point. You can provide reference material or describe a demonstrated workflow and ask Hermes to turn it into a skill. You can also correct the instructions when the process changes.
This is what learning means here: updating knowledge and procedures outside the model. It is not retraining the model's weights, and it is not a promise that each task makes the next one better.
Memory, skills and conversation history have different jobs
Hermes keeps compact memory for durable facts and user preferences. Skills hold procedures. Session search provides another way to consult earlier conversations without putting the entire history into every new prompt.
The distinction matters. A preference such as how you want reports formatted belongs in persistent context; a detailed reporting method belongs in a skill. A claim that the agent remembers something is only useful if it actually saved it.
Memory is bounded and curated, not a complete archive of everything you have read. If your main need is a searchable reference collection, our Recall knowledge-base overview addresses that different job. An agent's working memory should not be your only copy of important source material.
Official Desktop and Bot Mode
The official desktop app uses the same Hermes core, sessions, skills and memory as the CLI and gateway. Its documented interface includes multiple conversations, file browsing, attachments, visible tool activity and previews beside the chat.
Bot Mode adds named specialist profiles with their own roles, models, memory, skills and routines. You could separate a recurring research assistant from a coding assistant instead of asking one general conversation to carry both roles. Bot chats and groups provide a way to work with those specialists.
These persistent Bots are different from subagents. A subagent is a temporary worker delegated part of a task, with separate conversation context and a summary returned to its parent. Parallel research is a plausible use, but additional workers also use additional model resources.
Scheduled work, browser actions and connected tools
Hermes can schedule recurring jobs and deliver results through messaging channels such as Telegram, Discord or Slack. A daily research digest is a natural example. The gateway scheduler must be running when work is due; saving the schedule alone does not keep a sleeping laptop working.
Script-only jobs can run without an LLM turn when the task is deterministic. That is useful for checks that need a program, not a fresh piece of model reasoning.
Browser control handles interactions beyond search, while MCP connects external tools and services. Some integrations need their own credentials or paid accounts. A browser tool still has to deal with the site's login and interface; MCP support does not include a subscription to every connected service.
Get started with Hermes Agent for a repeatable workflow.
Hermes Agent vs Alternatives: Features and Cost Structure
OpenClaw is the closest comparison. Both support persistent memory, reusable skills, scheduled work and multiple models. OpenClaw also documents automatic skill learning and repair, so I would not choose Hermes on the assumption that OpenClaw cannot learn.
Hermes is worth a closer look if its skill workflow, Desktop and Bot Mode suit you, or if you want Portal and first-party Cloud hosting as optional conveniences. OpenClaw deserves equal attention if its Gateway and device-node approach match how you want to operate across machines.
Claude Code is a more focused candidate when most of your work is coding. It also has memory, skills and scheduled routines; its appeal is the development workflow, not the absence of those features elsewhere.
ChatGPT Work is the managed-service option to consider when you want research and deliverables without maintaining an open-source runtime. That trades away a different set of controls than switching between Hermes and OpenClaw.
Hermes Agent Pricing: Software, Portal Credits and Cloud
Free software, optional Portal subscription
Hermes Agent is free under the MIT license. You can use your own provider without subscribing to Nous Portal. Local inference is also supported, although suitable hardware and any external paid tools remain your responsibility.
Portal combines access to models and hosted tools through one account. Its advertised personal plans are:
Free provides access to free models with standard rate limits. Paid plans provide access to the broader catalog and hosted tools, with usage deducted from credits. The included credits are a spending allowance, not unlimited inference or unlimited browser sessions.
I would start with the provider arrangement you already understand. Portal is useful if consolidating model and tool billing solves a problem; a larger subscription is not necessary just to unlock the open-source agent.
Optional Nous-managed Cloud
Cloud keeps a hosted agent available without requiring your personal computer to remain on.
Running charges cover compute and storage, with compute metered by actual uptime. Model inference and tools are additional usage. Stopping the instance leaves a storage charge so its data can persist.
Portal credits can pay for eligible hosting, model and tool usage from the same balance. Do not add the full subscription price to every credit-funded usage charge as if they were independent cash bills. Your spending depends on consumption and any additional credit purchases.
Long sessions, repeated tool calls, subagents and scheduled jobs all affect consumption. There is no single monthly Hermes price that covers every workload.
Getting Started with Hermes Agent
Choose a model provider and decide whether work should run on your machine or a hosted instance. Then follow the official installation guide for the route you need.
Apple Silicon macOS, Windows and Linux/WSL2 have documented paths, but their installers and desktop packaging differ. Windows source installation and the packaged desktop app have different requirements. Intel Mac guidance is inconsistent across the current installation and platform pages, so I would not assume the main Mac download fits an Intel machine.
Start with one task whose result you can inspect, such as a saved research report. Add a reusable skill after the procedure is useful, and add scheduling when you have a reason to run it repeatedly. That keeps the initial setup tied to work you want done.
Verdict: Choose Hermes for Work You Want to Reuse
I would choose Hermes Agent for recurring research, development or operational work where retaining procedures matters. Its combination of editable skills, persistent context, specialist Bots and optional managed services gives you several ways to build that workflow.
Choose a focused coding tool if repositories dominate your work, or a managed assistant if operating the runtime is a burden you do not want. If you already run OpenClaw successfully, compare the day-to-day workflow before switching; a claim of exclusive self-learning is not a reason to migrate.
Try Hermes Agent with one recurring task.
FAQ
Does Hermes Agent require a Hermes model?
No. Hermes Agent is the runtime, not the model. It supports multiple providers and compatible local endpoints, so you choose the model separately.
Can Hermes keep working while my laptop is asleep?
Work that depends on that laptop cannot continue normally while it is unavailable. A hosted deployment can keep the necessary runtime and scheduler available independently of your laptop.
Is Hermes Agent completely free to use?
The software has no license fee. Hosted inference, paid tools and Cloud usage can cost money. A Portal subscription is optional, and its credits are not an unlimited-use plan.
Does learning change the underlying AI model?
The skills and memory described here store external instructions and context. They do not retrain the selected model, and their usefulness still depends on what the agent saves and how it uses it.
