Standard Compute
Hermes Agent
#3 of 21 · 8.4/10 editorialPersonal AgentOpen Source52% community win rate

Hermes Agent Review 2026: Ratings, Pros & Cons, Alternatives

Hermes Agent is Nous Research's open-source autonomous agent, released in February 2026 under the MIT license. Its defining feature is a built-in learning loop: after completing complex tasks it writes its own reusable skills, improves them with use, and builds persistent cross-session memory of you and your projects. It runs self-hosted (from a $5 VPS to a GPU cluster), works with 200+ models, and is reachable from the CLI or 20+ messaging platforms including Telegram, Discord, Slack, and WhatsApp.

Visit Hermes →Pricing: Free (MIT) / models via Standard ComputeUpdated 2026-10-16
What it costs to run Hermes

6 options compared, with prices and limits.

Best API for Hermes →See the Hermes integration →

Checked October 16, 2026 against the Hermes Agent docs and v0.22.2, the current release.

Hermes Agent browser automation

Hermes Agent has a full browser toolset: it can navigate websites, click and type on page elements, fill forms and extract information. Pages reach the model as accessibility-tree snapshots, and a vision tool can screenshot a page and analyze it.

By default it uses Browser Use mode, which writes and runs Python against a packaged headless Chromium, so nothing extra needs installing. You can instead attach your own Chrome, Brave, Chromium or Edge with /browser connect, run a local anti-detection browser (Camofox) or the lightweight Lightpanda engine, or use a cloud browser from Browser Use, Browserbase or Firecrawl. A paid Nous Portal plan includes a cloud browser through its tool gateway without separate API keys.

Hermes Agent memory and skills

Memory is deliberately small. MEMORY.md holds the agent's own notes (about 2,200 characters) and USER.md your preferences (about 1,375), both in ~/.hermes/memories/ and loaded at the start of each session. Past conversations are stored in SQLite with full-text search, so the agent can look up older sessions when it needs them. External memory providers such as Honcho or Mem0 can be plugged in.

Skills are instruction documents the agent loads on demand, stored in ~/.hermes/skills/ and compatible with the agentskills.io standard. Hermes ships a bundled catalog, you can install more from the Skills Hub, and the agent writes and patches its own skills after it solves new problems; each one becomes a slash command. A background curator moves unused agent-made skills to stale and then archived, so the catalog doesn't fill with near-duplicates.

Hermes Agent gateway platforms (chat apps)

The messaging gateway is one background process that connects Hermes to every chat app you configure: Telegram, Discord, Slack, WhatsApp, Signal, SMS, email, Microsoft Teams, Google Chat, Matrix, Mattermost, Home Assistant, iMessage through BlueBubbles, LINE, ntfy, Feishu/Lark, DingTalk, WeCom, Weixin, QQ and Yuanbao, among others. Set them up with hermes gateway setup.

The same process runs scheduled jobs and delivers voice replies on platforms that support them. Each chat is one long session that survives restarts, so Nous recommends running /new at natural breaks to keep the context short and let memory do its work.

Hermes Agent in 2026: what's new

Hermes Agent moves fast. v0.22.0, released October 13, 2026, opens the desktop app on a guided setup chat that runs on a free trial model, adds connectors that let the agent act in hosted apps such as Linear and Notion, runs local models with a one-click llama.cpp runtime, and moves installs to tested SemVer stable releases with signed desktop packages. v0.22.2 followed on October 15.

Hermes Agent v0.22: what changed →

Which model API to run Hermes Agent on

Hermes is free; the cost is the model calls, and an always-on agent makes a lot of them. On Standard Compute, the router sends each step to the cheapest model that can do it, so routine chat, memory and scheduled-job turns go to low-cost models and hard steps to frontier models. Providers are vetted for price, uptime and speed, with automatic fallback when one fails. Opt-in rules can keep a key on European or US providers, and we don't train on your prompts.

Best API for Hermes Agent →What Hermes Agent costs to run →Nous Portal pricing →

Sources: Hermes Agent docs: Browser · Hermes Agent docs: Persistent Memory · Hermes Agent docs: Skills System · Hermes Agent docs: Curator · Hermes Agent docs: Messaging Gateway · Hermes Agent releases on GitHub

How Hermes Scores

Output Quality
8.0/1049% votes
Autonomy
9.5/1066% votes
Reliability
8.5/1055% votes
Speed
8.0/1052% votes
Value
9.0/1053% votes
Ease of Use
7.5/1042% votes

Bars show editorial scores from hands-on testing. Percentages showHermes's live community win rate in head-to-head votes, shown once a category has 10 or more votes.

Hermes Pros, Cons, and Known Issues

Pros

  • The only major agent with a built-in learning loop: it writes its own reusable skills from experience and improves them with use
  • Persistent cross-session memory builds a deepening model of you and your projects
  • Reachable from the CLI or 20+ messaging platforms (Telegram, Discord, Slack, WhatsApp, Teams)
  • Model-agnostic across 200+ models; pair it with a flat-price plan via Standard Compute
  • MIT licensed, self-hosted, no telemetry: your data stays on your machine

Cons

  • Self-hosting means you own setup, updates, and security
  • Giving a long-running agent broad tool access requires careful permissioning
  • Young project (released February 2026), so best practices are still forming

Known issues users report

  • Auto-created skills vary in quality, so review what it teaches itself before relying on it
  • Persistent memory can accumulate stale facts over time and needs occasional pruning
  • Fast release cadence means breaking changes between versions

Key Features

Self-improving learning loop & skills
Persistent cross-session memory
CLI + 20+ messaging platforms
Web search & browser automation
Works with 200+ models
Self-hosted, MIT licensed, no telemetry

Runs on: macOS · Linux · Windows · VPS / GPU cluster / serverless|Best for: Power users who want a long-running personal agent that learns and compounds

Hermes vs the Alternatives: Vote in Any Matchup

HermesVSClaude CodeHermesVSCodexHermesVSGemini CLIHermesVSQwen CodeHermesVSOpenClawHermesVSKilo CodeHermesVSCursorHermesVSCopilotHermesVSClineHermesVSWindsurfHermesVSAiderHermesVSContinueHermesVSOpenCodeHermesVSPiHermesVSompHermesVSRoo CodeHermesVSDevinHermesVSAmpHermesVSJulesHermesVSAntigravity

Hermes Agent runs on Standard Compute

Standard Compute

Current frontier models Claude Fable 5.1 · GPT-6 Astra

More intelligence
per dollar.

Each step goes to the cheapest model that can do it, on providers we vet for price, uptime and speed, with automatic fallback when one fails.

Compliance
Opt-in rules keep a company, team or key on European or US providers
Privacy
We don’t train on your prompts or store them

Fixed monthly plans from $19 · cancel anytime · Hermes Agent setup →

7-day fair refund Free trial, no card Billing by Stripe3,800+ agent users

Hermes Agent: Frequently Asked Questions

Is Hermes Agent free?

Yes. Hermes Agent is MIT-licensed open source from Nous Research. You bring your own model access: an API key to any of 200+ supported models, or a flat-price plan via Standard Compute, which suits a long-running agent that makes constant model calls.

What is Hermes Agent best at?

Being a long-running personal agent that compounds: it remembers you across sessions, creates reusable skills from completed tasks, and is reachable from your messaging apps. It's built for ongoing delegation (research, automations, recurring workflows) rather than one-off coding sessions.

Hermes Agent vs OpenClaw: which self-hosted agent should I pick?

Both are MIT-licensed, self-hosted autonomous agents you reach through messaging apps. OpenClaw has the bigger ecosystem (100+ skills, massive community); Hermes has the learning loop: it writes and improves its own skills and builds persistent memory. Tinkerers who want breadth pick OpenClaw; users who want an agent that gets better over time pick Hermes. Vote on the matchup to add your take.

Can Hermes Agent do browser automation?

Yes. Hermes ships a browser toolset that navigates sites, clicks, fills forms and extracts information. The default Browser Use mode drives a packaged headless Chromium; you can also attach your own Chrome, Brave or Edge, or use a cloud browser (Browser Use, Browserbase, Firecrawl, or the Nous Portal tool gateway on a paid plan).

How do Hermes Agent memory and skills work?

Memory is two small files, MEMORY.md (the agent's notes) and USER.md (your profile), loaded at the start of every session, plus full-text search over past sessions. Skills are on-demand instruction documents in ~/.hermes/skills/, compatible with the agentskills.io standard; the agent writes new ones after solving problems, you can install more from the Skills Hub, and a background curator archives skills that go unused.

Which chat apps does the Hermes Agent gateway support?

The messaging gateway connects Hermes to Telegram, Discord, Slack, WhatsApp, Signal, SMS, email, Microsoft Teams, Google Chat, Matrix, Mattermost, Home Assistant, iMessage (via BlueBubbles), LINE, Feishu/Lark, DingTalk, WeCom, Weixin, QQ and more. One background process serves all configured platforms and also runs Hermes' scheduled jobs.

What's new in Hermes Agent in 2026?

Hermes Agent v0.22.0, released October 13, 2026, added a guided first-run setup on a free trial model, connectors for hosted apps such as Linear and Notion, one-click local models through llama.cpp, tested SemVer stable releases and signed desktop packages. Patch release v0.22.2 followed on October 15.