Hermes Agent

Hermes Agent is an open-source autonomous AI agent from Nous Research that you self-host on a VPS and chat with across 16 messaging platforms like Telegram, Discord, and Slack. It runs autonomous tasks with persistent memory and auto-generates reusable Python skills from your workflows.

Install takes 60 seconds with one bash script on Python 3.10+, and Hermes Agent tops the open-source AI agent category with 87,327 GitHub stars under MIT licence.

Features:

• •16 Messaging Platforms Hermes Agent handles Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Email, SMS, DingTalk, Feishu, WeCom, Mattermost, Home Assistant, Webhooks, iMessage via BlueBubbles, and WeChat. V0.9.0 added iMessage and WeChat, covering Apple and Chinese ecosystems.
• •3-Layer Persistent Memory Stores session data, long-term facts, and skill-specific knowledge. Recalls details across conversations without resets. Memory providers are pluggable so you can swap in Honcho, vector stores, or custom backends.
• •Auto-Generated Skills Turns completed tasks into reusable Python functions stored in persistent memory. DSPy and GEPA self-optimize prompts and skills over time without manual tuning.
• •Local Web Dashboard Run "hermes web" for a browser-based UI to monitor sessions, browse skills, configure gateways, and manage backups. Launched in v0.9.0 with no config files needed.
• •Fast Mode Priority-queue processing for OpenAI and Anthropic models including GPT-5.4 and Claude. Toggle /fast to cut latency on time-sensitive tasks without switching providers.
• •Background Process Monitoring Set watch_patterns to track long-running builds, deployments, or tests without polling. Auto-notified when a pattern matches or the process finishes.
• •Worktree Parallelism And Browser Use Run multiple git tasks in isolated parallel worktrees. Built-in Browser Use integration handles browser automation tasks without leaving the agent.
• •Remote Execution Backends Deploys tasks to Docker, Modal, SSH, Singularity, or Daytona. MCP OAuth 2.1 connects to any MCP server your stack exposes via the CLI.
• •Pluggable Context And Model Switching Swap context engines and memory providers via hermes plugins. The CLI /model command switches live among 400+ models from OpenRouter, Ollama, Nous Portal, Hugging Face, Google AI Studio, xAI Grok, Anthropic, OpenAI, Xiaomi MiMo, and Qwen.

Pricing:

• •Hermes Agent $0/mo, free under MIT licence with no subscription fee ever.
• •Self-Hosted VPS ~$5/mo for a CPU-only server running Ollama for local models; total hosting lands $5-$20/mo depending on instance size and traffic.
• •Nous Portal LLMs $0/mo free tier of ~10k tokens/month for personal use; OpenRouter pay-as-you-go runs DeepSeek-V3 at ~$0.30 per million tokens.
• •BYOK $0/mo from Hermes; bring your own API keys for OpenAI, Anthropic, Grok, Gemini, and others, with LLM costs billed directly by your chosen provider.

Limitations:

• •Always-on server required Hermes Agent cannot run on a laptop that sleeps or a phone. You need a dedicated VPS or always-on machine. Budget $5-$20/month for a reliable host.
• •Gateway config is technical The 60-second bash install is smooth, but connecting Telegram, Discord, or Slack requires creating bot tokens and editing YAML config files. Not a click-and-go setup for non-technical users.
• •No native iOS app Termux runs Hermes Agent natively on Android. iOS users need a Mac with BlueBubbles for iMessage access, a two-device setup with its own complexity.
• •LLM provider costs are separate Hermes Agent is free software, but you pay your chosen AI provider. OpenRouter costs ~$0.30 per million tokens for DeepSeek-V3; OpenAI and Anthropic cost significantly more at heavy usage.
• •Smaller mainstream community than SaaS agents The 87,327-star repo is large, but troubleshooting outside GitHub issues is thinner than for hosted SaaS agents like ChatGPT or Claude. Self-service answers take longer.