Hermes Agent
Nous Research · Self-Hosted Autonomous Agent with Persistent Memory
An open-source autonomous agent from Nous Research that runs on your own server, accumulates memory across every session, and creates reusable skills from experience. Built for developers and technical teams who want a persistent, private, model-agnostic AI agent they fully control.
Best for
Developers and technical power users who want a persistent, self-hosted AI agent with compounding memory and full control over their infrastructure and model choice
Not ideal for
Non-technical users or anyone who wants a ready-to-use cloud AI assistant without server setup and API key management
Who it's for
Developers and technical teams who want a self-hosted autonomous agent with persistent memory and no vendor lock-in
Capabilities
- Persistent layered memory stored as markdown files on your own machine — grows across every session and project
- Autonomous skill creation: agent builds and refines new capabilities from its own experience without manual programming
- Model-agnostic: any OpenRouter model, Nous Portal (300+ models), or direct connections to Anthropic, OpenAI, Google, DeepSeek, AWS Bedrock, Azure AI Foundry, Ollama, vLLM, LM Studio, and any OpenAI-compatible endpoint
- Operates across 20+ platforms: Telegram, Discord, Slack, WhatsApp, Signal, Email, Microsoft Teams, SMS, Matrix, Google Chat, Home Assistant, iMessage, and CLI
- OpenAI-compatible API server for frontends like Open WebUI and LobeChat
- Built-in cron scheduler for natural-language-defined recurring automations
- Parallel subagents for handling multiple concurrent tasks or conversations
- Web search, browser automation, vision processing, and image generation
- MCP (Model Context Protocol) server integration
- Built-in OpenClaw migration: `hermes claw migrate` auto-imports settings, memories, skills, and API keys, with dry-run preview and selective import
- Hermes Desktop installer for macOS (Apple Silicon) and Windows, plus shell and PowerShell installers
- Skills compatible with the agentskills.io open standard
Limitations
- Requires self-hosting and technical setup — not suitable for non-technical users
- No built-in model — you must supply your own API keys or run local models; this adds cost and configuration overhead
- Memory and skill quality build over time; limited value out-of-the-box before the memory layer accumulates meaningful context
- The agent is free and MIT-licensed; Nous Portal (300+ hosted models plus the Tool Gateway) has a free tier, with paid tiers from $20/month
Use cases
- Running a persistent AI agent on your own VPS that retains full context across every project indefinitely
- Automating recurring tasks via natural-language cron schedules without writing custom scripts
- Deploying a multi-platform AI presence across Telegram, Discord, and Slack from a single self-hosted server
- Privacy-sensitive workflows where sending prompts and context to cloud providers is not acceptable
- Power users who want a fully customizable autonomous agent without any vendor dependency
Our take
Hermes Agent's real value is persistence — not memory that can be toggled on, but an agent that accumulates knowledge across every session and stores it as readable markdown files on your own machine. The compounding effect is genuine: the longer you run it, the more contextual its behavior becomes because it builds up your workflows, preferences, and knowledge structures over time. The cost is equally real: you manage your own server, your own API keys, and all the operational overhead of a self-hosted system. Users who want a cloud chatbot with a memory toggle will find this overkill. Users who want their AI agent to learn their business deeply over months, run scheduled automations, and keep all data on their own infrastructure will find this is closer to what they actually needed.
Who should use it
Developers and technical teams running long-horizon projects who want an AI agent that improves with use and keeps all data and memory on their own infrastructure.
Who should skip it
Non-technical users, or anyone who wants a working AI assistant without managing server infrastructure, API keys, and ongoing maintenance overhead.
Strengths
- Compounding memory: knowledge accumulates across every session and project, improving responses over time
- Fully self-hosted: all data — conversations, memory, skills — stays on your own machine or server
- Model-agnostic: works with any LLM provider or local models via Ollama, vLLM, or LM Studio
- Multi-platform deployment: one installation operates across Telegram, Discord, Slack, WhatsApp, and CLI
- Free and open source under MIT license — no licensing fees for the agent itself
- Skill creation: agent builds and refines reusable capabilities from its own experience
Weaknesses
- Requires server management, API key configuration, and comfort with the terminal — not accessible to non-technical users
- Cold-start problem: memory value builds over weeks of use, not immediately useful out-of-the-box
- No proprietary model included — LLM API costs are your responsibility and vary by provider
- Higher maintenance burden than cloud AI tools — you own updates, uptime, and server health
Hermes Agent pricing
Hermes Agent
Free
- Open source under MIT license
- No subscription fee for the agent itself
- Self-host on Linux, macOS, Windows, or Android (Termux)
- Bring your own LLM provider API key
With local models (Ollama)
$0 API cost
- Run any Ollama-compatible model locally
- No API tokens consumed
- All inference stays on your own hardware
- Setup requires a GPU or capable CPU
With cloud LLMs
Pay-per-token (varies)
- Anthropic, OpenAI, Google, DeepSeek, and other providers
- Free models available on some providers such as OpenRouter
- Billed directly by your chosen provider
Nous Portal (optional)
Free / $20 / $100 / $200 per month
Billed monthly
- 300+ hosted models behind one login
- Tool Gateway for web, image, TTS, and browser tools
- Paid tiers include matching usage credits ($22, $110, $220)
Note: The agent itself is always free. Your only cost is the LLM inference you choose to connect. Local models via Ollama have no API cost. Cloud API pricing is set by your chosen provider and billed directly to you.
Technical specs
Modalities
Text, Image (vision), Image generation, Voice (TTS), Web browsing, Browser automation
Available models
Where Hermes Agent excels
A personal knowledge base that grows with use
Memory accumulates as readable markdown files on your machine — Hermes builds a persistent understanding of your projects, preferences, and working patterns that carries into every session.
Recurring automations defined in plain language
The built-in cron scheduler lets you set up daily reports, reminders, or data pulls without writing scripts. You describe the task in plain language; Hermes runs it on schedule.
Multi-platform AI presence from one server
One Hermes installation operates across Telegram, Discord, Slack, WhatsApp, and a CLI simultaneously — useful for small teams spread across different communication tools.
Privacy-first AI workflows
All conversation history, memory, and skills stay on your own server. The only external calls are to the LLM API you choose — and with Ollama, even those stay local.
Hermes Agent vs. competitors
Hermes Agent vs. ChatGPT (with memory)
ChatGPT's memory stores summaries in OpenAI's cloud and works within platform limits. Hermes Agent stores complete, layered memory as local files on your own machine with no platform-side retention. ChatGPT is more accessible; Hermes gives data ownership and true cross-session persistence.
Hermes Agent vs. Claude
Claude has strong instruction following and up to 1M-token context windows but resets between sessions. Hermes can wrap Claude (or any model) with persistent cross-session memory on your own infrastructure. Claude is better for individual high-quality tasks; Hermes is better for ongoing, evolving workflows that compound over time.
Hermes Agent vs. n8n / Make
n8n and Make are visual workflow automation platforms. Hermes is an autonomous agent with adaptive memory and skill-building. Hermes can trigger automations via cron, but if you need structured multi-step automation with clear logic flows and no server management, n8n or Make are more appropriate.
Frequently asked questions
Is Hermes Agent free?
Yes. The agent itself is free and open source under the MIT license. You only pay for the LLM API you connect to — or use local models via Ollama at no API cost.
Can I use Hermes Agent on Windows?
Yes. Native Windows 10/11 is a Tier 1 platform (install with PowerShell or Hermes Desktop); a few features are not available natively, and WSL2 is also supported.
What makes Hermes Agent different from ChatGPT with memory?
ChatGPT's memory is stored on OpenAI's servers and works within platform limits. Hermes Agent stores all memory as readable markdown files on your own machine — no cloud retention, no platform dependency. It also creates and refines its own skills from experience, which ChatGPT does not do.
Does Hermes Agent require a specific LLM?
No. It is fully model-agnostic. Connect any LLM via OpenRouter, use direct provider APIs from Anthropic, OpenAI, Google, and others, or run local models via Ollama or vLLM at no API cost.
How technical is the setup?
You can install with the Hermes Desktop installer (macOS Apple Silicon and Windows), a shell script (Linux, macOS, WSL2, Termux), or PowerShell (Windows), then configure your preferred LLM provider. A basic VPS or local machine is sufficient. Comfort with the terminal and server management is still expected for self-hosted, always-on setups.
Integrations & fit
Alternatives to consider
OpenClaw vs Hermes Agent — full comparison →Anthropic
Developers and writers who need the most accurate code generation, precise instruction following, and the ability to reason across very large documents
FreemiumOpenAI
Knowledge workers who need one AI subscription covering writing, research, coding, image generation, and team collaboration
Freemium
AgentScope
Technical users and small teams who want a self-hosted assistant in DingTalk, Feishu, WeCom, WeChat, or QQ, offline use with small local models, and built-in tool guards with an OS-level shell sandbox
Free
TinyHumans
Individuals and solo operators on macOS, Windows, or Linux who want a desktop agent that keeps a local, readable memory of their email, chat, and work apps and can act across them, without wiring API keys or running a server
FreemiumAbout Hermes Agent
Hermes Agent runs persistently on infrastructure you own — a laptop, VPS, or GPU cluster — and stores layered memory as readable markdown files on your machine. Unlike session-based AI assistants that reset after each conversation, Hermes accumulates knowledge, creates new skills from its own experience, and deepens its model of your work over time. It is model-agnostic: connect any LLM via OpenRouter or Nous Portal (300+ models behind one login), local models via Ollama, or direct provider APIs from Anthropic, OpenAI, Google, and others. It also exposes an OpenAI-compatible API server, so frontends like Open WebUI and LobeChat can talk to it. It operates across 20+ messaging platforms including Telegram, Discord, Slack, WhatsApp, SMS, Matrix, Google Chat, and a CLI, and runs on Linux, macOS, native Windows 10/11, and Android (Termux). Best suited for developers who want full infrastructure ownership, data privacy, and long-running autonomous workflows — not a suitable choice if you want a zero-setup cloud assistant.
Are you the founder? Claim this listing →