Nous Research

Hermes Agent

Name: Hermes Agent
Author: Nous Research

Self-improving open-source agent with a built-in learning loop: it creates skills from experience, keeps persistent memory across sessions, and runs from the terminal or messaging gateways (Telegram, Discord, Slack, WhatsApp) on seven execution backends from a $5 VPS to serverless sandboxes. The #1 coding app on OpenRouter by routed tokens.

CLICloudWebActiveOpen sourceBYOKMCP Yes

n/a

Attributed output

No signal we can detect yet

n/aAttributed outputNo signal we can detect yetCommits and PRs on public GitHub provably produced by this agent. Exact if it is visible through its own bot or workflow, an estimate if measured through attribution it declares (these can swing 30 to 40 percent between measurement runs at multi-million scale), and a floor either way.

#33

Index rank

of 44 tracked agents

#33Index rankof 44 tracked agents

n/a

Business signal

No sourced business metric

n/aBusiness signalNo sourced business metric

n/a

Entry price

No published pricing

n/aEntry priceNo published pricing

236,244

GitHub stars

NousResearch/hermes-agent

236,244GitHub starsNousResearch/hermes-agent

Not published

npm demand/wk

Public npm counter

Not publishednpm demand/wkPublic npm counterWeekly package downloads as an install-demand floor; adjusted when the vendor's main install channel is no longer npm.

Market

Developer output

No signal we can detect yet

Hermes Agent works through the developer's own identity and leaves nothing public we can currently attribute to it on its output, so its public footprint is invisible \u2014 not absent, and not a sign of low usage. Agents in this group are ordered by their sourced business and traction signals instead.

Index rank: #33 of 44
Demand share: No public npm package
Business signal: No sourced business metric
GitHub share: 15.1% of tracked stars
npm channel: No public npm package

All sourced signals (2)Expand

Signal	Value	Type	Confidence
OpenRouter coding rank	#1 coding app, ~2.47T tokens in the current window (~2.5x Claude Code's routed volume)	Public metric	High
v0.20 'Herald'	Voice mode with barge-in and A2A shipped ~Aug 8, 2026; signed compiled-binary releases announced Aug 23	Official	High

Pricing And Economics

Expand

Free, MIT-licensed software; you pay your model provider (Nous Portal, OpenRouter, or any BYOK endpoint).

The agent itself is free (MIT); model spend goes to whichever provider is configured.
Serverless backends (Modal, Daytona) hibernate when idle, so a persistent agent can cost near-zero between sessions.

Product

Providers

Nous PortalOpenRouterOpenAIAnthropicSelf-hosted endpoints

Model families

Any via provider config (Hermes, Claude, GPT, open-weights)

Capabilities

Skill creation from experiencePersistent cross-session memoryMessaging gateways (20+ platforms)SubagentsScheduled automationsMCPagentskills.io standardSeven terminal backends

Tradeoffs

Learning loop and memory make runs less reproducible by design
Fast-moving release cadence; binary signing only announced Aug 2026
Distinct from the Hermes model family despite the shared name