Skip to content
Agentic Levels
Start hereLevelsTracksPromptsAbout
Get my levelGet my level

  • Start hereYour first useful task with AI.
  • Level quiz8 questions. No AI needed.
  • Fluency ScoreYour Level and a score, from your own AI.

  • LevelsThe 11 levels, in order.
  • LessonsShort lessons, filter by level.
  • TracksGuided paths by tool, role or goal.
  • Set up your AI tool
  • Starter packFree files that help AI remember.

  • PromptsReady prompts to copy and adapt.
  • ChatGPT vs Claude vs GeminiPick the right one for the job.
  • How-tos
  • ReferenceCommands for Claude Code, Codex, Antigravity.
  • ResourcesHand-picked reading and tools.

  • Your progress
  • Your next moveAn interview that picks your next skill.
  • Your next lessonsThe next lessons at your level.
  • Updates

  • Pricing
  • About
  • FAQ & Feedback
  • Work with us
  • How we handle your data
Sign inPreferences
AboutPricingHow-tosPromptsStarter packMy stackHow we handle your data

© 2026 Fuentes Studio

PrivacyTermsRefundsAccessibilityAI content policyCredits
yourCouncil
Ready to help
✦

What do you want to understand?

Ask anything about what you're learning.

←How-to guides
  1. Resources
  2. ›How-to guides

How to use Hermes Agent

Nous Research's self-improving AI agent with cross-session memory, multi-platform messaging, serverless deployment, and a closed skill-learning loop.

Levels 5 to 7On GitHub

Most agent frameworks require you to host infrastructure, manually configure each platform integration, and rebuild context every session. Hermes Agent runs on a $5 VPS or serverless (Modal, Daytona, Vercel Sandbox), connects to Telegram, Discord, Slack, WhatsApp, Signal, and CLI from a single install, and builds a cross-session memory via FTS5 full-text search. Its SONA self-learning loop creates skills from experience and improves them during use. So the agent gets better the more you run it. With 200+ LLM providers via OpenRouter, 40+ built-in tools, and a cron scheduler, Hermes is a production-deployable agent you run once and operate indefinitely.

Before you start

Python 3.11+ and pip. A gateway API key for your preferred messaging platform (Telegram is the simplest to start). An LLM provider key, OpenRouter covers 200+ models with one key. Clone the repo and run the installer.

  1. 1Install and start the gateway
  2. 2Switch the model to match your task
  3. 3Schedule an automated task with hermes cron
  4. 4Search cross-session memory with FTS5
  5. 5Use an execution backend for isolated task runs

Steps

01

Install and start the gateway

Clone the repo and run the installer script. Choose your gateway. Telegram is the fastest to set up: create a bot via BotFather, get the token, and pass it to hermes gateway. The agent is immediately reachable from your phone or any Telegram client.

Install and start Telegram gateway
git clone https://github.com/NousResearch/hermes-agent cd hermes-agent pip install -r requirements.txt # Set your API keys: export OPENROUTER_API_KEY=[YOUR_OPENROUTER_KEY] export TELEGRAM_TOKEN=[YOUR_BOT_TOKEN] # Start the gateway: python -m hermes gateway telegram
[YOUR_OPENROUTER_KEY]OpenRouter API key (covers 200+ models) e.g. sk-or-...
[YOUR_BOT_TOKEN]Telegram bot token from BotFather e.g. 1234567890:ABC...
02

Switch the model to match your task

Hermes routes to any of 200+ models via OpenRouter. Use a fast cheap model for quick tasks and a powerful one for complex reasoning. The switch takes one command and persists for the session.

Switch model
# In your terminal or via gateway: hermes model [MODEL_NAME] # Examples: hermes model anthropic/claude-sonnet-5.5 hermes model openai/gpt-4o hermes model meta-llama/llama-3.3-70b-instruct
[MODEL_NAME]Model identifier (OpenRouter format) e.g. anthropic/claude-sonnet-5.5
03

Schedule an automated task with hermes cron

Hermes has a built-in cron scheduler. Define tasks that run on a schedule without you present, daily reports, monitoring checks, or periodic data pulls. Tasks run in the background and surface results to your gateway.

Schedule a recurring task
hermes cron add --schedule "[CRON_SCHEDULE]" --task "[TASK_DESCRIPTION]"
[CRON_SCHEDULE]Cron expression e.g. 0 9 * * 1-5
[TASK_DESCRIPTION]What to run e.g. Check the status of all GitHub Actions workflows in my repos and report any failures
04

Search cross-session memory with FTS5

Hermes stores every conversation turn in SQLite with FTS5 full-text search. Ask the agent to recall past context. It searches across all sessions, not just the current one. This is how SONA self-learning works: patterns from past sessions inform current behavior.

Search past sessions
# Via gateway or CLI: hermes memory search "[SEARCH_QUERY]"
[SEARCH_QUERY]What to search for in past sessions e.g. deployment steps for my Next.js app
05

Use an execution backend for isolated task runs

For tasks needing isolation, running untrusted code, spinning up a fresh environment, or deploying to serverless, Hermes supports 7 backends: local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox. Select the backend per task without changing agent code.

Run a task in Modal serverless
# Set backend in config or per-invocation: hermes run --backend modal "[TASK_DESCRIPTION]"
[TASK_DESCRIPTION]Task to run in isolation e.g. Run the test suite and return the output, fresh environment, no local deps

How do you know it worked?

  • The gateway is running and you can send messages to the agent from Telegram, Discord, or CLI.
  • hermes model switches the active LLM and the next response uses it.
  • A cron task fires on schedule and delivers its result to your gateway.
  • hermes memory search returns relevant results from past sessions.
  • At least one task runs in an isolated backend (Docker or Modal) without affecting your local environment.

If something looks off

Gateway tokens are credentials, treat them as secrets
Your Telegram bot token or Discord webhook gives anyone who has it access to your running agent. Never commit them to version control. Use environment variables or a secrets manager.
SONA self-learning changes behavior over time
The self-learning loop improves the agent by extracting skills from successful runs. If the agent starts behaving unexpectedly, review recently created skills with hermes skills list and delete any that are wrong.
Cron tasks run unattended, scope them defensively
Scheduled tasks have no human in the loop. Make sure each cron task has explicit boundaries: what it reads, what it writes, and what it should never do. A task with broad permissions running unattended is a surface for accidental changes.

Skills & Commands Reference

Core Hermes Agent commands for gateway, model routing, cron, memory, and execution backends.

Command / SkillWhat it doesWhen to use it
hermes gatewayStart the messaging gateway for a platform: telegram, discord, slack, whatsapp, signal, or cliFirst command, choose your primary interface before doing anything else
hermes modelSwitch active LLM provider/model, supports 200+ via OpenRouterChanging from a fast/cheap model to a powerful one for complex reasoning tasks
hermes cronadvancedSchedule automated tasks using cron expressions, runs unattended in the backgroundRecurring workflows: daily reports, monitoring checks, periodic data pulls
SONA self-learningadvancedClosed skill-learning loop that creates and improves skills from successful task runsActive automatically, review hermes skills list periodically to audit learned behaviors
hermes memory searchFTS5 full-text search across all past session conversationsRecalling decisions, steps, or context from previous sessions without re-explaining
Execution backendsadvanced7 backends (local/Docker/SSH/Singularity/Modal/Daytona/Vercel Sandbox) selected per taskIsolated execution for untrusted code, fresh environments, or serverless deployment

What you'll do, 5 steps

  1. 1Install and start the gateway
  2. 2Switch the model to match your task
  3. 3Schedule an automated task with hermes cron
  4. 4Search cross-session memory with FTS5
  5. 5Use an execution backend for isolated task runs

Was this helpful?