Hermes Agent
Officially listedAn open-source, self-hosted always-on autonomous agent runtime from Nous Research, model-agnostic by design.
Hermes Agent
Hermes Agent is an MIT-licensed, self-hosted always-on autonomous agent runtime from Nous Research — model-agnostic, bring your own API key. It is not a coding copilot hanging off an IDE, nor a chat shell for a single model; it is the "body" handling memory, skills, tools, and cross-platform dispatch, growing more capable the longer it runs. Target users are developers and small technical teams with server/Docker experience who value data sovereignty.
Core Capabilities
- Persistent memory: Autonomously maintains MEMORY.md and USER.md, stores full history in a database with FTS5 retrieval, understanding you better over time
- Self-built skills: After completing complex tasks it automatically writes reusable SKILL documents, continuing to self-improve on the next invocation
- Message gateway: A single process covering Telegram/Discord/Slack plus DingTalk/Feishu/WeCom/WeChat and 20+ platforms total
- Model-agnostic: Works with any OpenAI endpoint plus Anthropic/Gemini, and can also run fully offline on local Ollama
- Two form factors: Terminal CLI plus native desktop apps (macOS/Windows/Linux), with 60+ built-in tools, MCP/ACP, and scheduled jobs
Use Cases Suits developers and small technical teams with container and bash experience who want always-on automation (scheduled briefings, backups, cross-platform research); not for non-technical users who want out-of-the-box with zero ops.
Unique Advantages Where OpenClaw wins on ecosystem breadth, Hermes takes the "auto-written skills + cross-session persistent memory" learning-loop route — working memory out of the box is the core reason it has captured roughly 30% of migrating users, and it is MIT-licensed, zero-telemetry, with all data local.
Editor's Verdict One of the most noteworthy open-source agents of 2026, with strong momentum and capital backing, but still at v0.x with a very large open issue count and about 73% fixed token overhead — heavy use can cost over a thousand dollars. Also be alert: the copycat site hermes-agent.ai misuses the official brand to sell FlyHermes, and malicious ads impersonate the official one; trust only hermes-agent.nousresearch.com and never enter through search ads or run suspicious scripts.
Pricing
### 💰 定价模式：开源免费（软件本体）+ 可选托管订阅 **起步价**：免费（MIT 开源，自托管零软件费用） #### 主要方案 - **Hermes Agent 本体**：$0 - MIT 许可、零遥测，仅自付 LLM API 与主机（最低约 $5/月 VPS） - **Nous Portal Free**：$0/月 - 仅免费模型、标准限速 - **Nous Portal Plus**：$20/月 - $22 额度、200+ 模型与托管工具 - **Nous Portal Super / Ultra**：$100 / $200 每月 - $110 / $220 额度，rollover 上限更高 #### 试用/其他信息 注意：Nous Portal 是「可选」的托管模型与算力订阅，并非 Hermes Agent 软件的收费——软件本体完全免费。第三方实测重度自动化 API 成本可达 $800–1,500/月，token 固定开销约 73%。 — Visit website
FAQ
Is Hermes Agent free?
The software itself is completely free: MIT open source, self-hosted, zero telemetry — you only pay your own LLM API and hosting costs. The optional Nous Portal hosted subscription is a model-credit and compute service, not a fee for the Hermes Agent software.
What can Hermes Agent do?
It is a resident autonomous agent: persistent memory across sessions, automatic creation and iteration of skills, reachability over 20+ messaging platforms (Telegram, DingTalk, WeChat, etc.), plus 60+ built-in tools and natural-language scheduled tasks, suited to long-term automation.
How does Hermes Agent differ from the Hermes open models?
Hermes Agent is a self-hosted agent runtime (the application layer) handling memory, skills, and scheduling; Hermes 3/4 are models (the brain) from the same lab. The agent is model-agnostic — it can use Hermes models, or switch to Claude, GPT, or local Ollama.
How does Hermes Agent differ from OpenClaw?
Both are MIT and self-hosted. Hermes Agent centers on auto-written skills plus cross-session persistent memory as a learning loop, with working memory out of the box; OpenClaw has a broader ecosystem and integrations with more community skills, and about 30% of community users have migrated from OpenClaw to Hermes.
How does Hermes Agent differ from Claude Code?
Hermes Agent is a self-hosted, model-agnostic resident autonomous agent with persistent cross-project memory and connections to 20+ messaging platforms; Claude Code is a closed-source terminal/IDE coding copilot locked to its own models, geared toward single-session development.
Do I have to bring my own API key to run Hermes Agent?
Not necessarily, but most setups need model credits. Hermes Agent is model-agnostic: you can supply any provider's API key, use the official Nous Portal, or run fully offline with local Ollama/vLLM — local models are completely free.
How do I confirm the real Hermes Agent site and avoid copycats?
Trust only the official domain hermes-agent.nousresearch.com and install only via the official install.sh/install.ps1. The copycat site hermes-agent.ai misuses the brand to sell FlyHermes, and malicious ads impersonate the official one — never enter through search ads or execute suspicious scripts.