Hermes Agent
Self-hosted personal AI agent from Nous Research that lives inside Telegram, Discord, Slack and other messaging apps, with persistent memory and sandboxed task execution.
🔗 Visit Hermes Agent
Description
Most AI assistants live in a single chat window you have to remember to open. Hermes Agent instead moves into the apps you already have open all day — Telegram, Discord, Slack, WhatsApp, Signal, email — so it can remember context across all of them, run tasks in the background, and check in when something needs your attention.
Hermes Agent is an open-source (MIT license), self-hosted AI agent from Nous Research that connects to multiple messaging platforms simultaneously, keeps persistent memory and auto-generated skills across sessions, and runs tasks — browsing, reporting, briefings, backups, multi-step workflows — through one of five sandboxing backends (local, Docker, SSH, Singularity, or Modal) for isolation. It supports multi-model reasoning with vision and image generation, and gives access to 300+ models through Nous's own portal on paid tiers, alongside a free tier. Since launching in July 2025, it has become one of the most-starred projects in this space, with 238k+ GitHub stars, 48k+ forks, and very active daily development.
💬 Our review
The short version: Hermes Agent's GitHub numbers (238k+ stars) are genuinely startling for a project barely a year old, and the cross-platform, sandboxed, persistent-memory design is technically ambitious — but that same scope means real setup complexity, and a wide-open 37k+ issue count is worth noting before you commit.
Against narrower personal-agent tools like NanoClaw, which focuses on containerized messaging-app agents specifically, Hermes Agent is more ambitious in scope: multiple sandbox backends, vision and image generation, and access to 300+ models via Nous's portal rather than a single provider. Against pointing Claude Code or a GPT-4-powered assistant directly at a task, Hermes Agent's advantage is persistence and cross-platform reach — it remembers context across messaging apps and runs unattended, rather than living in one session. The tradeoff is exactly what you'd expect from an extremely popular, fast-moving open-source project: setup and sandbox configuration take real effort, and the open issue count suggests plenty of rough edges are still being worked through in public. If you want a genuinely persistent, multi-platform personal agent and don't mind some setup friction, it's one of the most capable open options available right now; if you want something that works in five minutes, a simpler single-platform tool will get you there faster.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Palier gratuit disponible ; paliers payants Plus/Super/Ultra avec crédits mensuels et accès à 300+ modèles via le portail Nous
Pros
open source sous licence MIT avec 238k+ étoiles GitHub et 48k+ forks
support multi-plateforme desktop (macOS, Windows, Linux) et messagerie (Telegram, Discord, Slack, WhatsApp, Signal)
mémoire persistante avec compétences auto-générées entre les sessions
raisonnement multi-modèle avec vision et génération d'images
cinq backends de sandboxing (local, Docker, SSH, Singularity, Modal)
développement très actif (dernier commit le jour du contrôle)
Cons
projet jeune (lancé en juillet 2025) avec un nombre élevé de tickets ouverts (37 818)
transparence tarifaire limitée, détails des paliers nécessitent une inscription
complexité de configuration et du sandboxing potentiellement élevée pour les utilisateurs non techniques