Hermes is Nous Research's open-source (MIT) AI agent with a built-in learning loop: it creates and refines its own skills from experience and builds a persistent model of the user across sessions. Available via terminal, desktop app or the Nous Portal, with Business and Enterprise (self-hosted or cloud) tiers for teams. Pick it when you want an agent that improves with use, not a fixed feature set.
Other good fits
A team wanting a shared agent with collective skill development (Hermes Business) or a self-hosted, sovereign deployment (Enterprise).
+ Pros
Read pros
Hermes is Nous Research's open-source AI agent, built around a learning loop: it creates skills from experience, improves them during use, nudges itself to persist knowledge, and builds a deepening model of who you are across sessions, rather than behaving identically every time you open it. It's MIT-licensed and freely available, accessible through a terminal install, native desktop apps (Mac, Windows, Linux), or the cloud-based Nous Portal.
Pick it when…
You want an agent whose usefulness compounds over time as it builds and refines its own skills from your actual use of it.
You want the flexibility to run it locally (terminal, desktop) or through a cloud portal, depending on the task.
A team wants a shared agent (Hermes Business) with collective skill development, or full sovereignty over deployment (Hermes Enterprise, self-hosted or cloud).
− Cons
Read cons
An agent that builds a persistent model of you and changes its own skills is less predictable than a static tool; if you need consistent, auditable behaviour every time, that learning loop is a real trade-off to weigh. Business and Enterprise pricing wasn't published on the pages checked for this review; confirm current terms directly with Nous Research before planning a team deployment.
Pick something else when…
You want predictable, unchanging behaviour every time. A static agent framework, like the Claude Agent SDK, gives you that instead.
You want agent orchestration defined explicitly in code you control.CrewAI is a code-first alternative without a self-modifying skill system.
You want a visual, no-code way to build agent workflows.Dify or Flowise fit that better.
Avoid if
You need a simple, single-purpose tool; Hermes's learning-loop framing adds complexity you may not need for a narrow task.
Workflow
Install via terminal or desktop app, or start through the Nous Portal → give it real tasks and let it build skills from that experience → periodically check what it's learned and persisted about your patterns, so its behaviour stays predictable rather than surprising → for a team, consider Hermes Business for shared skill development, or Enterprise for a self-hosted, sovereign deployment.
Cost
Hermes itself is open source (MIT) and free to run locally via terminal or desktop app; you pay only for the compute or model access it uses underneath. The Nous Portal (cloud-based) and Business/Enterprise tiers likely carry their own pricing not detailed on the pages checked for this review; check current terms directly before committing a team to a paid tier. The ongoing effort, beyond typical agent maintenance, is periodically reviewing what it's learned and persisted about you or your team, since a self-improving agent's behaviour can drift over time in ways a static tool's won't.
Restrictions
Hermes Business and Enterprise pricing and terms weren't detailed on the pages checked for this review; contact Nous Research directly for current team pricing. Self-hosted Enterprise deployment needs your own infrastructure and the operational responsibility that comes with it.
First project
Documentation-based suggestion — not run here.
Give Hermes a recurring, real task over several sessions, for example drafting a weekly status update from the same kind of input each time, and observe whether its output genuinely improves or adapts to your preferences across sessions, not just within one.
Done when you can point to a concrete way its behaviour changed for the better after repeated use, or conclude honestly that it didn't for your specific task.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on Nous Research's homepage, read on 2026-09-23. No Hermes instance was installed or run for this review; Business and Enterprise pricing was not published on the page checked.
Nous Research homepage · Sep 23, 2026 — Hermes agent overview, MIT license, terminal/desktop/portal access, Business and Enterprise tiers.
Claude Agent SDK overview · Sep 23, 2026 — Comparison alternative with predictable, static agent behaviour.
CrewAI homepage · Sep 23, 2026 — Comparison alternative for code-defined agent orchestration.
Further review notes & alternatives
Alternatives and trade-offs
Claude Agent SDK wins when you want predictable, static agent behaviour embedded in your own application.
CrewAI wins when you want agent orchestration explicitly defined in code, not a self-modifying skill system.
Dify wins when a visual, no-code builder fits your team better than a terminal or desktop agent.
Keep persistent project memory across sessions with a self-hosted agent.
Schedule tasks through a CLI or familiar messaging applications.
Create reusable skills from solutions you want to repeat.
Nous Research · free · self-host
Container image: nousresearch/hermes-agent:latest
persistent memory
autonomous scheduling
skill generation
messaging integration
self-hosted deployment
local models
docker container
cron tasks
automate
talk-think
write-code
open-source
View sourced imageHermes Agent sessions dashboard from the official documentation repository. This is the documented web administration interface with example sessions.
Nous Research — Hermes Agent dashboard documentation
Owner-approved use of public official interface screenshots, as recorded in public-visuals-batch-1.json, 13 September 2026. Copyright retained by Nous Research and Hermes Agent contributors; attribution retained. Editorial product illustration; no open licence asserted. Converted on 15 October 2026 to a WebP (quality 80) for faster page loads; the original image remains in the media library.
