Last updated: 2026-09-27
🪁

Hermes

Long-horizon · Persistent memory · Autonomous tasks · MCP tools
MIT licensed v0.21.5 stable Curated memory + session search MCP tool support Writes its own skills

Hermes Agent is an open-source agent from Nous Research that runs on your own machine or server. You talk to it from a terminal, its desktop app, or Telegram, Discord and 20-odd other chat apps. What sets it apart is that it learns: after a task, a background review saves what it learned about you and your setup and writes or refines skills it can reuse next time. Add scheduled jobs, MCP tools and a web dashboard, and it can do a lot of work while you're away.

Current release: v0.21.5 (tag v2026.9.24) — update with hermes update

v0.21.0 "Pantheon" (Aug 31) added Bot Mode in the desktop app (named bots in shared group chats), hermes peer bot-to-bot messaging, cron jobs that remember previous runs, live steering of subagents, an MCP command centre, and Meta's Muse Spark as a built-in provider. v0.21.2 fixed a state.db bug in 0.21.0–0.21.1 that could report healthy databases as corrupt, and closed a leak between profiles sharing one gateway. If you run 0.21.0 or 0.21.1, update now (details). Earlier in the summer: v0.19 made smart approvals the default and added Bitwarden/1Password secret sources; v0.20 added voice with barge-in, wake words, A2A and outbound webhooks. Install is a one-line script or desktop package, not pip; see the Quick Start.

⚡
Quick Start — 15 Minutes
Install with the desktop package or one-line script, pick a provider and a 64K+ context model, run your first chat, and keep the gateway running as a service.
Live
🛠
Write Your Own Skills
Self-writing skills are Hermes's whole identity. How the self-improvement loop works, a copy-paste prompt to have the agent author a skill safely, and the review checklist before you let one persist.
LiveNew
🔐
Security & Hardening
Hermes' real controls: allowlists and DM pairing, approval modes and deny rules, a Docker terminal backend, run limits, dashboard login, secret sources and hermes security audit, with a checklist.
LiveNew
📊
Web Dashboard (localhost:9119)
What each page does (chat, config, keys, sessions, cron, skills, MCP, channels) and how to reach it safely over an SSH tunnel.
LiveNew
✈️
Channel Setup: Telegram
Create with QR or BotFather, TELEGRAM_ALLOWED_USERS, the gateway, a home channel for scheduled jobs, and mention-only groups.
LiveNew
💸
Best Free Models
Which free OpenRouter, Nous-portal, and Gemini models actually clear the 64K-context + reliable-tool-use bar to drive the agent — and the two-model trick to dodge rate limits.
LiveNew
🧠
How Memory Works
MEMORY.md and USER.md, the frozen snapshot, session search over state.db, write approval, /journey, and external providers like Honcho and Mem0.
Live
🗓
Scheduled Jobs, Goals & Kanban
hermes cron schedules and delivery, script-only watchdogs, jobs that remember, /goal loops, and the multi-agent kanban board.
Live
🔌
MCP Servers
The Nous-reviewed catalog, mcp_servers in config.yaml for stdio and HTTP servers, per-server tool filtering, OAuth, lazy start, and writing your own.
Live
⚖️
Hermes vs OpenClaw
What each actually does today: memory, learning, scheduling, channels and safety defaults, plus hermes claw migrate and running both.
Live
🔐
VPS Install — Side-by-Side with Kilo Code
Hardened Hetzner Ubuntu 24.04 install with per-user isolation and zero public ports beyond SSH, first tested on v0.11 and updated for the current installer. Every gotcha with the fix.
LiveNew
💬
Discord Gateway — The Definitive Setup
Five silent failure modes the Discord gateway can hit. Developer Portal config, the systemd linger + bus-socket fix, the auto_thread trap, channel-permission overrides, multi-channel project layout.
LiveNew
🛠️
Troubleshooting & FAQ
Every error message and weird behavior we hit during a real April 2026 Hermes + Kilo install on Hetzner, with the fix that actually worked. SSH, isolation, install, runtime, Discord, systemd, Kilo, and operational FAQ.
LiveNew
❓
Hermes FAQ
Top Hermes Agent questions answered: why it needs a 64K-context model, how to fix tool-use failures, memory tuning, model selection, and when Hermes beats OpenClaw. Updated weekly from community discussion.
Live

At a Glance

LicenseMIT (fully free)
InstallDesktop package (macOS Apple Silicon, Windows) or curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
Current versionv0.21.5 (2026-09-24). Update with hermes update
RequiresGit, curl, tar; the installer brings its own Python 3.14 and Node
Config~/.hermes/config.yaml, with secrets in ~/.hermes/.env
MemoryMEMORY.md (2,200 chars) + USER.md (1,375 chars), plus full-text search over past sessions
Schedulinghermes cron: "every monday 9am", cron expressions, script-only jobs
ToolsBuilt-in terminal, files, web and browser; MCP servers; self-written skills
ChannelsTelegram, Discord, Slack, WhatsApp, Signal, email, SMS, Matrix, iMessage (BlueBubbles) and more
ModelAny model with 64K+ context. Claude Sonnet 5 or GPT-6 Sol are strong defaults

Hermes Use Cases — Long-Running & Memory-Enabled

Hermes is built for always-on agents that learn over time. These are the canonical use cases.

Hermes Troubleshooting

Hermes Security

You run Hermes with your own keys, so channel allowlists, command approvals and secret hygiene matter most. Start with the Hermes hardening guide.

Related on This Site

Official sources

Checked against the Hermes Agent documentation (v0.21.5) on 2026-09-27. If anything here disagrees with these pages, trust the vendor docs.

📬 Weekly Digest — In Your Inbox

One email a week: top news, releases, and our deepest new guide. No spam. Same content via RSS if you prefer.