How do I get started?

Download Hermes Desktop for macOS or Windows, or run the terminal install script. Follow setup to connect a model provider, then start a conversation. For an always-on hosted agent, choose Deploy in the cloud.

Does my agent keep its data between sessions?

Yes. Hermes stores conversations, memories, and skills so you can return to your work later. Keep a backup of your Hermes data directory when moving to another computer.

What happens when I stop my agent?

Stopping a local agent ends its running process but does not delete its saved conversations, memories, or skills. Start it again to continue. Scheduled tasks will not run while it is stopped; hosted agents are managed separately in Nous Portal.

How do I chat with my agent?

Use Hermes Desktop or the terminal, or connect a messaging platform such as Telegram, Discord, Slack, WhatsApp, Signal, or email. The docs walk through configuring each integration through the messaging gateway.

How does the self-improving skill loop actually work?

Hermes uses a closed learning loop: it curates memory with periodic nudges, autonomously writes new skills from solved tasks, and refines those skills during reuse. Recall is backed by FTS5 full-text search across sessions with LLM summarization, plus Honcho dialectic user modeling to build a profile of you. Skills follow the open agentskills.io standard, so they are portable and shareable.

Can I sandbox the agent so it never touches my host machine?

Yes. Hermes ships six terminal backends — local, Docker, SSH, Daytona, Modal, and Vercel Sandbox — plus plugin backends like Singularity. Docker and Singularity add container hardening and namespace isolation, while Daytona and Modal hibernate when idle so serverless environments cost nearly nothing. Pair that with command approval rules and authorization policies for a hardened setup.

Directory/Automation/AI Assistants/Hermes Agent

Hermes Agent

Hermes Agent is Nous Research's open-source autonomous AI agent with a built-in learning loop. It creates skills from experience, refines them during use, and maintains persistent memory (FTS5 cross-session recall plus Honcho dialectic user modeling) so it gets more capable the longer it runs. A messaging gateway puts the same agent and memory on Telegram, Discord, Slack, WhatsApp, Signal, Email, Matrix and 20+ other surfaces. Under the hood it ships 60+ tools, MCP server support, web search, browser automation, vision, image generation and TTS, plus isolated subagents and Programmatic Tool Calling via execute_code. Six terminal sandbox backends (local, Docker, SSH, Daytona, Modal, Vercel Sandbox) let it hibernate on serverless infra when idle. Cron-based scheduling runs reports and backups unattended. It is a terminal-native, framework-agnostic agent aimed at engineers and researchers who want to self-host — on a $5 VPS, a GPU cluster or serverless — and plug in their own model prov

Automation

Visit website

Added on

Nov 4, 2026

MIT license20+ platformsPersistent memorySkill creationSubagentsWeb browsingMCP supportVoice modeCloud hosting

Product information

Everything worth knowing about Hermes Agent

The complete picture — from everyday features to the technical detail builders and IT folks dig for.

Core features

What it actually does

One Agent, Every Surface

Lives in Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Mattermost, Email, SMS, Teams, Google Chat and more — 20+ platforms through a single gateway with one shared memory.

Persistent Memory

Remembers your projects across sessions, auto-generates skills from what it solved, and nudges itself to persist knowledge instead of starting from zero every time.

Natural-Language Scheduling

Built-in cron automation for reports, backups and briefings, delivered to any platform and running unattended through the gateway.

Subagents & Delegation

Spawns isolated subagents with their own conversations, terminals and Python RPC scripts so parallel workstreams don't burn your main context window.

Full Web Control

Web search, extraction, browser automation, vision, image generation and text-to-speech — all available as tools, bundled via one Nous Portal subscription.

Technical capabilities

Closed Learning Loop

Agent-curated memory with periodic nudges, autonomous skill creation, skill self-improvement during use, FTS5 cross-session recall with LLM summarization, and Honcho dialectic user modeling.

Isolated Sandboxing

6 built-in terminal backends (local, Docker, SSH, Daytona, Modal, Vercel Sandbox) plus plugin backends like Singularity, with container hardening and namespace isolation; serverless backends hibernate when idle.

Must watch videos

52:51

Tutorial52:51

Nous Office Hours - December 2025

Recording of our December 2025 office hours, featuring some audience Q/A and discussion about Psyche. Watch our current training runs here: https://psyche.network/runs Read our Psyche announcement here: https://nousresearch.com/nous-psyche GitHub: https://github.com/PsycheFoundation/psyche

23:02

Tutorial23:02

RL Environments Explained

Roger from Nous Research explaining the Atropos project, RL Environments inside of it and talking about our hackathon in May 2025. Check out the Atropos repo here and experiment! https://github.com/nousresearch/atropos

1:04:15

Tutorial1:04:15

Nous Office Hours- July 24

Recording of our office hours on July 24, featuring some audience Q/A and discussion about Psyche. Watch our current training run here: https://psyche.network/runs/consilience-40b-1/0 Read our Psyche announcement here: https://nousresearch.com/nous-psyche/ GitHub: https://github.com/PsycheFoundation/psyche

Who uses it

Real use cases, no hype

Always-On Personal Agent

Run Hermes on a $5 VPS or serverless backend and chat with it from Telegram while it works on a cloud VM you never SSH into.

Scheduled Automations

Natural-language cron for daily briefings, backups, and reports delivered straight to Slack, email, or Discord.

Parallel Agent Pipelines

Spawn isolated subagents with their own terminals and Python RPC scripts to run workstreams in parallel without burning context.

Multi-Platform Support Bot

Bot Mode gives named specialist bots their own model, memory, and skills, coordinating via @mentions in group chats.

Research & Trajectory Export

Batch processing, trajectory export, and RL training with Atropos for teams building on agent data.

Web Research & Scraping

Search, extract, browse, and screenshot with vision and browser automation bundled through one Portal subscription.

Safe Code Execution

Run untrusted agent code inside Docker, Singularity, or Modal with namespace isolation and container hardening.

What's great

Free and open source under MIT — zero vendor lock-in
One gateway, 20+ chat platforms (Telegram, Discord, Slack, Signal, email)
Persistent memory that survives sessions, restarts, and reboots
60-second install via a single curl or PowerShell command

Technical strengths

Six terminal backends — Docker, SSH, Daytona, Modal, Vercel Sandbox plus local
Self-authored skills with FTS5 cross-session recall and LLM summarization
execute_code collapses multi-step pipelines into a single inference call

Where it falls short

You pay for model tokens — free tier is free models only
Not an IDE copilot; no inline autocomplete in your editor
Self-hosting means you own uptime, updates, and data backups
Surface area is huge — 60+ tools and 20+ platforms is a lot to learn

Technical limitations

Unpruned memory and skill stores bloat context and inflate token costs
Autonomous tool use demands approval rules and container hardening
Hosted tools (web search, image gen, TTS, browser) are paywalled behind Portal

For technical folks

The deep-dive specs

License

MIT (open source)

Terminal backends

6 built-in: local, Docker, SSH, Daytona, Modal, Vercel Sandbox

Plugin backends

Singularity and others

Memory recall

FTS5 cross-session recall with LLM summarization

User modeling

Honcho dialectic user modeling

Built-in tools

60+

Messaging platforms

20+

Skill compatibility

Open standard, agentskills.io compatible

Model providers

Nous Portal, OpenRouter, OpenAI, or any endpoint

Extensibility

MCP server integration

FAQ

Includes technical Q&A

Hermes Agent is free and open source under the MIT license. Model providers and optional hosted services have their own pricing. You can use your own provider or connect to Nous Portal for model and tool credits; cloud hosting is optional.

Traffic Insights

Monthly Visits

—

November 2026

Growth Rate

vs last month

Dominance

in Automation

Top Country

—

Top Countries Breakdown

—

Growth Trend

Stable

Traffic has remained stable.

Honest pricing

No sneaky tiers, no “contact sales”

Here's who each plan is actually for — and where the hidden charges might hit.

Free

Individuals trying it out

Best for trying Hermes on free models; free models only; no hosted tools; no cloud hosting; standard rate limits

Plus

Most picked

$20/mo

Individuals

Everyday personal use; shared usage; pay-as-you-go top-ups; free and paid models; hosted tools (web search
images
speech
cloud browser); cloud hosting 4 GB RAM
25 GB disk; high rate limits

Super

$100/mo

Researchers & coders

Frequent research and coding; 5× Plus usage; pay-as-you-go top-ups; free and paid models; hosted tools; cloud hosting 4 GB RAM
100 GB disk; high rate limits

Ultra

$200/mo

Heavy individual users