Index / Hermes Agent · updated Oct 2, 2026
Hermes Agent
NousResearch/hermes-agent · healthy · rank 2 of 43 by stars
A self-improving personal AI agent from Nous Research with a built-in learning loop that creates and refines skills from experience. It runs anywhere from a $5 VPS to a GPU cluster and reaches you via Telegram, Discord, Slack, WhatsApp, Signal, or CLI.
Facts
Repository
Runtime
Posture
Community
Security breakdown
Composite 72 / 100 · how these are scored
higher is safer
higher is safer
higher is safer
higher is safer
higher is riskier
Model access
Read from the repository, not written by a model · 79 files examined
10 providers · runs locally · gateway support
Pinned models
claude-opus-4.6 released 2026-02-04 — 8 months old (from the public model catalogue)
Pin last edited 14 days ago
Several defaults are pinned; the project picks one per provider rather than shipping a single default.
Evidence
Decision
Why choose Hermes Agent over OpenClaw?
Why choose this
- Built-in learning loop with autonomous skill creation and self-improvement
- Native multi-channel messaging gateway (Telegram, Discord, Slack, WhatsApp, Signal)
- Cross-session memory with FTS5 search and Honcho user modeling
Tradeoffs
- Python runtime is heavier and slower to boot than leaner alternatives
- Broad feature surface increases operational complexity and failure modes
- Security posture relies on path allowlists rather than deep sandboxing
Best fit
- Users wanting a persistent personal agent with long-term memory
- People who chat with their agent over Telegram/Discord while it runs on a VPS
- Tinkerers who want model-agnostic, no-lock-in agent infrastructure
Avoid if
- You need strict sandboxing for untrusted tool execution
- You want a minimal, single-purpose CLI without a gateway stack
- You are cost-sensitive and prone to runaway token spend on paid models
Strong evidence from an active README, frequent recent commits, a large Reddit community, and third-party comparisons. Security and sandboxing depth are inferred from commit-level fixes rather than a formal security model, so those scores carry moderate uncertainty.
AI layer reviewed Sep 21, 2026 · how this is written
Star activity
250,623 stars today
Overview
Hermes Agent is Nous Research's flagship open-source personal AI agent, marketed as 'the agent that grows with you.' Its defining feature is a closed learning loop: the agent autonomously creates skills after complex tasks, improves those skills during use, nudges itself to persist knowledge, and searches its own past conversations via FTS5 with LLM summarization. It also integrates Honcho for dialectic user modeling, building a deepening profile of the user across sessions, and is compatible with the agentskills.io open standard.
Architecturally, Hermes is model-agnostic — it works with Nous Portal, OpenRouter, OpenAI, or any custom endpoint, switchable at runtime with hermes model. A single gateway process connects the agent to Telegram, Discord, Slack, WhatsApp, Signal, and CLI, with voice memo transcription and cross-platform conversation continuity. It ships a full terminal UI, a desktop Electron app, scheduled automations, and deployment options ranging from a $5 VPS to serverless or GPU clusters. Recent commits show heavy investment in desktop stability, credential resolution correctness, and safe file-path handling for chat attachments.
Compared to OpenClaw, Hermes positions itself as the user-facing assistant with memory and self-improvement as first-class features, even shipping a hermes claw migrate command for OpenClaw refugees. It is reportedly the #1 tracked app on OpenRouter by token volume, and its subreddit shows an active community sharing real-world automations, cost lessons, and third-party clients for iOS and Android.