Hermes Agent: the AI agent that grows with you
Hermes Agent is an open-source AI agent launched by Nous Research on February 25, 2026, which racked up over 57,000 GitHub stars in six weeks. Its promise: an agent that grows with you through persistent memory and auto-generated skills. Unlike static agents, Hermes runs a self-improving loop: it observes failures, detects patterns, and creates new skills to handle future cases. It supports the agentskills.io open standard, making it interoperable with OpenClaw. Hermes works with 15+ messaging platforms and ships under the MIT license. It's the pick of developers who want a truly adaptive agent, not just another LLM wrapper.
For who
Experienced developers, AI product teams and researchers who want an evolving agent with long-term memory and self-learning.
Input
Taches en langage naturel, contexte utilisateur, historique de conversation. Hermes utilise cette entree pour enrichir sa memoire et declencher la generation de skills.
steps (6)
Install Hermes
infoClone the Hermes repo, install Python dependencies, and configure the memory backend (SQLite or Postgres).
Set up persistent memory
infoInitialize the vector store for long-term memory. Hermes supports pgvector, Qdrant and Chroma.
Enable the self-improvement loop
infoTurn on the skill-forge module that detects recurring failures and auto-generates new agentskills.io skills.
Connect your messaging
infoLink Hermes to Telegram, Signal, Discord or any of the 15+ supported platforms.
Onboard the agent
infoGive Hermes a set of real tasks for a few days to feed its memory and let skills emerge.
Monitor the evolution
infoRegularly inspect auto-generated skills and memory. Correct wrong skills manually to guide learning.
Output
Reponses contextualisees tenant compte des interactions passees, nouveaux skills generes et actions executees sur les plateformes connectees.
Example
Input
After several weeks of usage: "As usual, prepare my weekly report for the marketing team from Stripe metrics and LinkedIn posts."
Output
Hermes recognizes the pattern from its memory, reuses the auto-generated weekly-marketing-report skill from previous weeks, pulls Stripe numbers, scrapes LinkedIn posts, and sends the formatted report directly to Slack.
Customization
| Parameter | Description | Default |
|---|---|---|
| Backend memoire | Choix du vector store (pgvector, Qdrant, Chroma) pour la memoire a long terme. | pgvector |
| Generation de skills | Activer/desactiver la boucle de generation automatique de skills. | Active |
| Plateformes connectees | Messageries branchees a l'agent parmi les 15+ supportees. | Telegram, Signal |