Hermes Agent

Hermes Agent is an open-source AI agent framework by Nous Research. It exposes an OpenAI-compatible API with cross-session memory, allowing it to recall context from previous conversations. Completed workflows are saved as reusable skill files. It can browse web pages and run scheduled tasks.

How it works

This template deploys Hermes Agent in gateway mode, which starts an HTTP API server on your Zeabur instance. The API follows the OpenAI chat completions format, so any compatible client can connect to it.

The agent itself does not run LLM models locally — it forwards requests to external providers (OpenRouter, OpenAI, Anthropic, or any OpenAI-compatible endpoint you configure).

How to use

Option 1: Connect a Chat UI

Deploy a frontend like NextChat or Open WebUI on Zeabur, then configure it to point to your Hermes Agent:

• API Base URL: https://<your-domain>
• Model name: hermes-agent
• API Key: leave empty (not required by default)

Option 2: Call the API directly

# Send a message
curl https://<your-domain>/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model": "hermes-agent", "messages": [{"role": "user", "content": "Hello"}]}'

# List available models
curl https://<your-domain>/v1/models

# Health check
curl https://<your-domain>/health

Option 3: Connect messaging platforms

Add bot tokens in Service > Variables after deployment:

• TELEGRAM_BOT_TOKEN — Telegram integration
• SLACK_BOT_TOKEN + SLACK_SIGNING_SECRET — Slack integration
• DISCORD_BOT_TOKEN — Discord integration

Key features

• Cross-session memory — conversation history is stored locally and searchable across sessions
• Skill files — completed multi-step workflows are saved as reusable skill definitions
• Web access — built-in browser for reading web pages; optional Exa/Firecrawl integration for search
• Task scheduling — cron-based scheduler for recurring tasks
• Multi-provider — supports OpenRouter, OpenAI, Anthropic, and custom OpenAI-compatible endpoints

Configuration

Required: At least one LLM provider API key.

Variable	Purpose
OPENROUTER_API_KEY	Access multiple models via OpenRouter (recommended for getting started)
OPENAI_API_KEY	Use OpenAI models directly
ANTHROPIC_API_KEY	Use Anthropic Claude models directly
LLM_MODEL	Override the default model selection
EXA_API_KEY	Enable Exa web search
TELEGRAM_BOT_TOKEN	Enable Telegram bot

Set optional variables in Zeabur Service > Variables after deployment.

Notes

• RAM: 1 GB minimum, 2-4 GB recommended. Browser automation (Playwright) requires at least 2 GB. The agent may be terminated by the system if memory is insufficient.
• First deployment takes a few minutes due to image size (~2.4 GB compressed). Subsequent restarts pull from cache and start in seconds.
• No GPU needed — the agent calls external LLM APIs over HTTP.
• All state (memory, skills, config) survives restarts.

License

MIT — GitHub