Hermes Agent

Overview

Hermes Agent is a self-hosted, open-source AI agent framework built by Nous Research that gets smarter the longer you use it. Released in 2026 under the MIT license (exact release timing should be verified from the repository), it runs on your own server and learns from every task it completes — writing reusable skill files, storing outcomes in persistent memory, and adjusting its approach for next time.

What sets it apart from typical agent frameworks — including autonomous AI agents that rely on cloud infrastructure — is the closed learning loop. Solve a problem once, and Hermes turns the solution into a Markdown skill that improves with each subsequent use. It supports any OpenAI-compatible model provider, connects to Telegram, Discord, Slack, WhatsApp, Signal, and Email from a single gateway, and runs comfortably on hardware as modest as a $5 VPS. With tens of thousands of GitHub stars and 40+ bundled skills (exact count may vary over time), it has quickly become one of the most actively adopted open-source agent projects of 2026.

Key Features

• Self-improving skill loop — Every completed task generates a reusable Markdown skill file that the agent refines over time, so repeated workflows get faster and more accurate without manual configuration.

• Persistent cross-session memory — Stores all conversations and context in a SQLite database with FTS5 full-text indexing, similar in spirit to dedicated AI memory systems, with fast full-text search (latency may vary by dataset size and hardware) so the agent can retrieve past work efficiently.

• Multi-platform gateway — Access the same agent from Telegram, Discord, Slack, WhatsApp, Signal, Email, or CLI through a single process, with conversations synced across platforms.

• Model-agnostic architecture — Switch between Nous Portal, OpenRouter (200+ models), OpenAI, MiniMax, Kimi/Moonshot, or your own endpoint with zero code changes and no vendor lock-in.

• Subagent delegation — Spawns isolated subagents with independent conversations and terminal access to parallelize complex tasks without context-loss across pipelines.

• Scheduled automations — Built-in cron scheduling lets you set up recurring reports, backups, and briefings in natural language that run unattended through the gateway.

How to Get Started

Getting Hermes Agent running takes two commands. Install the CLI with the official script:

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

After installation, reload your shell and run hermes setup to configure your model provider and preferred messaging platforms. The setup wizard walks through API key entry, gateway selection (Telegram, Discord, Slack, etc.), and terminal backend choice.

Hermes supports six terminal backends — local, Docker, SSH, Daytona, Singularity, and Modal — so you can match your deployment to your infrastructure. For most users, the local backend works out of the box. Teams needing container isolation can switch to Docker or Singularity with a single config flag.

Community & Ecosystem

With over 35,000 GitHub stars, Hermes Agent has built a substantial community since its February 2026 launch. The awesome-hermes-agent repository curates community-contributed skills, integrations, and deployment guides.

The framework ships with 40+ bundled skills covering MLOps, GitHub workflows, research pipelines, web scraping, and code execution. Community members actively publish additional skills through the Skills Hub, and the agent's self-improving architecture means any user can generate new skills simply by completing tasks — no plugin development required.

MCP (Model Context Protocol) support arrived in v0.6.0, allowing Hermes to connect to any MCP server for extended tool capabilities. The current release is v0.8.0, which added typed SDK workflows and improved subagent coordination.

Pricing & Plans

Hermes Agent is open source under the MIT license and does not include official paid tiers. However, actual usage costs depend on your infrastructure and model providers.

Your only costs are infrastructure and model API usage:

• Self-hosting: Can run on low-cost VPS setups (often around $5/month), a GPU cluster, or serverless platforms like Modal and Daytona (actual costs vary by usage and provider)
• Model API: Depends on your chosen provider — use free local models via Ollama, or pay per-token through OpenRouter, OpenAI, or Nous Portal
• No telemetry or tracking: Your data stays on your machine with no cloud lock-in

Best For

• Developers who want a self-hosted AI agent — as an alternative to tools like Claude Code — that they fully control and can extend with custom skills
• DevOps engineers automating recurring infrastructure tasks across Slack, Discord, or Telegram
• Researchers and ML engineers needing a flexible agent framework that supports any model provider
• Privacy-conscious teams that require all data to remain on their own servers
• Open-source enthusiasts looking for a community-backed alternative to proprietary agent platforms

FAQ