Hermes Agent: The Self-Improving AI Agent

What Even Is Hermes Agent?

The AI agent landscape has grown crowded, but most tools fall into two camps: chatbot wrappers over a single API, or coding copilots tethered to an IDE. Hermes Agent, built by Nous Research (the lab behind the Hermes, Nomos, and Psyche model families), is neither. It's an open-source, autonomous agent designed to get more capable the longer it runs.

The headline feature is its closed learning loop — Hermes creates skills from experience, improves them during use, nudges itself to persist knowledge, and builds a deepening model of who you are across sessions. It's not just a tool that answers questions; it's an agent that learns, remembers, and adapts.

And it doesn't live on your laptop. Deploy it on a $5 VPS, a GPU cluster, or serverless infrastructure like Daytona or Modal that costs nearly nothing when idle. Message it from Telegram while it works on a cloud VM you never SSH into.

---

The Learning Loop: Why Hermes Gets Smarter Over Time

Most AI tools start every session from scratch. Hermes doesn't. Its architecture revolves around three interconnected systems:

Persistent Memory

Hermes maintains bounded, curated memory across sessions via MEMORY.md and USER.md files. It remembers your preferences, your projects, your environment, and things it has learned. This isn't a naive "dump everything into context" approach — memory is actively curated and summarized using FTS5 cross-session recall with LLM summarization.

Skills System

Skills are on-demand knowledge documents the agent loads when relevant. They follow a progressive disclosure pattern (minimizing token usage) and are compatible with the agentskills.io open standard. The agent doesn't just consume skills — it creates them from its own experience and improves them during subsequent use.

hermes skills search kubernetes
hermes skills install openai/skills/k8s

Browse community-contributed skills through the Skills Hub, or let Hermes generate its own from repeated workflows.

Dialectic User Modeling

Through Honcho integration, Hermes builds a progressively deeper understanding of who you are — not just what you ask, but how you work, what you prefer, and what context matters most to you.

---

Features That Actually Matter

Hermes ships with an absurd number of capabilities. Here are the ones worth paying attention to:

Core

• Tools & Toolsets — Web search, terminal execution, file editing, memory, delegation, and more. Tools are organized into logical toolsets that can be enabled or disabled per platform.
• Context Files — Hermes auto-discovers project context files (.hermes.md, AGENTS.md, CLAUDE.md, SOUL.md, .cursorrules) to shape its behavior per project.
• Context References — Type @ followed by a reference to inject files, folders, git diffs, and URLs directly into your messages.
• Checkpoints — Automatic directory snapshots before file changes, with /rollback to undo mistakes.

Automation

• Scheduled Tasks (Cron) — Schedule tasks with natural language or cron expressions. Jobs can attach skills, deliver results to any connected platform, and support pause/resume/edit.
• Subagent Delegation — The delegate_task tool spawns isolated child agents with restricted toolsets and their own terminal sessions. Run up to 3 concurrent subagents by default (configurable) for parallel workstreams.
• Code Execution — The execute_code tool lets the agent write Python scripts that call Hermes tools programmatically, collapsing multi-step workflows into a single LLM turn.
• Batch Processing — Run Hermes across thousands of prompts in parallel, generating structured ShareGPT-format trajectory data for training or evaluation.

Media & Web

• Voice Mode — Full voice interaction across CLI and messaging platforms. Talk via microphone, hear spoken replies, and join Discord voice channels for live conversations.
• Browser Automation — Multiple backends: Browserbase cloud, Browser Use cloud, local Chrome via CDP, or local Chromium. Navigate, fill forms, and extract information.
• Vision & Image Paste — Paste images from your clipboard into the CLI for analysis with any vision-capable model.
• Image Generation — Generate images via FAL.ai with eight supported models including FLUX 2, GPT-Image 1.5, and Recraft V4 Pro.

Integrations

• MCP Support — Connect to any MCP server via stdio or HTTP transport. Access external tools from GitHub, databases, and internal APIs without writing native Hermes tools.
• Provider Routing — Fine-grained control over which AI providers handle your requests. Optimize for cost, speed, or quality.
• Fallback Providers — Automatic failover to backup LLM providers when your primary model encounters errors.
• API Server — Expose Hermes as an OpenAI-compatible HTTP endpoint. Connect Open WebUI, LobeChat, LibreChat, or any compatible frontend.
• IDE Integration (ACP) — Use Hermes inside VS Code, Zed, and JetBrains. Chat, tool activity, file diffs, and terminal commands render directly in your editor.

---

Runs Everywhere, Talks Everywhere

Hermes supports six terminal backends: local, Docker, SSH, Daytona, Singularity, and Modal. Daytona and Modal offer serverless persistence — your environment hibernates when idle and costs nearly nothing.

On the messaging side, Hermes connects to 15+ platforms from a single gateway:

• Telegram, Discord, Slack, WhatsApp, Signal
• Matrix, Mattermost, Email, SMS
• DingTalk, Feishu, WeCom, BlueBubbles
• Home Assistant

Set it up with a single command:

hermes gateway setup

Your Hermes instance becomes a bot you can message from your phone while it works on infrastructure you never touch.

---

Getting Started in Under 5 Minutes

1. Install

# Linux / macOS / WSL2 / Android (Termux)
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
source ~/.zshrc  # or source ~/.bashrc

2. Choose a Provider

hermes model

This walks you through provider selection interactively. Hermes works with Nous Portal, OpenRouter, OpenAI, or any compatible endpoint. The only hard requirement: at least 64K tokens of context.

3. Start Chatting

hermes       # classic CLI
hermes --tui # modern TUI (recommended)

Try something concrete to verify everything works:

❯ Summarize this repo in 5 bullets and tell me what the main entrypoint is.

4. Resume Sessions

hermes --continue  # or hermes -c

Sessions persist automatically. Pick up where you left off.

5. Add Layers as Needed

Once basic chat works, layer on features:

hermes gateway setup                      # messaging bots
hermes config set terminal.backend docker  # sandboxed execution
hermes skills search kubernetes            # install skills
pip install "hermes-agent[voice]"          # voice mode

---

Customization & Personality

Hermes is deeply customizable. SOUL.md is the primary identity file — the first thing in the system prompt — and you can swap in built-in or custom personality presets per session with /personality. Skins and themes let you customize the CLI's visual presentation, including banner colors, spinner animations, and branding.

The plugin system supports three types: general plugins (tools/hooks), memory providers (cross-session knowledge), and context engines (alternative context management). All managed through:

hermes plugins

---

Why Hermes Is Different

There are plenty of AI agents out there. What makes Hermes worth a closer look:

1. It learns from itself. The skill creation → improvement → memory loop is built into the architecture, not bolted on. The agent literally gets better the more you use it.
2. It's infrastructure-agnostic. Six terminal backends, 15+ messaging platforms, serverless deployment. It runs where you need it, not where it's convenient for the vendor.
3. It's built by model trainers. Nous Research trains the models that power many open-source AI applications. Hermes Agent is built with deep understanding of what LLMs can and can't do well.
4. It's open source. Full source on GitHub. No lock-in, no black boxes.
5. It's research-ready. Batch processing, trajectory export, and RL training with Atropos make it a first-class tool for AI researchers, not just end users.

If you're tired of AI tools that forget everything between sessions and can't operate beyond the confines of a single IDE, Hermes Agent is worth your time.