Hermes Agent: the self-hosted agent that learns
Hermes Agent is the answer to a question a lot of Central Coast businesses have but rarely ask out loud: what do I do if I cannot — or will not — send my data to OpenAI, Anthropic, or Notion?
Released by Nous Research in February 2026. Crossed 60,000 stars on GitHub in two months — the fastest-growing open-source AI agent project of the year. Latest release (v0.13, May 7) is called the Tenacity Release and adds a durable multi-agent task board.
What makes it different
Three things set Hermes apart from the closed-source alternatives:
One — it lives on your server. A $20-a-month VPS, your own datacenter, a Raspberry Pi in your back office. Your data does not leave your network unless you tell it to.
Two — it learns. Hermes is the only mainstream agent with a built-in learning loop. It creates skills from experience, refines them during use, persists what it learns about your business, and searches its own past conversations. Over weeks of running, it becomes specifically yours.
Three — it lives where you already chat. Native support for Telegram, Discord, Slack, WhatsApp, Signal, Email, and CLI. Start a conversation on one platform, pick it up on another. The agent is the same person across surfaces.
What v0.13 added
The Tenacity Release shipped two features worth flagging. Kanban: a durable multi-agent task board with heartbeat, reclaim, zombie detection, auto-block on incomplete exit, per-task retries, and hallucination recovery. Translation: agents can now actually finish multi-step jobs without losing track of where they were. And /goal — a slash command that locks the agent onto a specific target across many conversation turns. This is the open-source flavor of Plan Mode and Claude's goal-seeking pattern.
When to choose Hermes
Choose Hermes if any of the following are true for your business:
You handle regulated data — medical, legal, financial — and cannot risk it leaving your network. You have customers who specifically asked you not to use US AI vendors. You want to own the agent rather than rent it. You want to run agent workflows over chat tools your team already uses (Slack, Telegram, Signal) without bouncing them out to a separate web app.
Choose one of the closed-source alternatives (Notion, AgentKit, Claude) if: setup time matters more than data sovereignty, you do not have an IT person willing to maintain a server, or your workflow already lives inside one of the major platforms.
Setup reality
You will need: a server (any VPS, $5–$20/month), a model provider (you can run a local open-weights model or call out to a hosted one — Hermes supports both), and an afternoon. The repo at github.com/NousResearch/hermes-agent has the install path. For a business with a technical partner, this is a Saturday. For one without, this is where we typically help.
Hermes is the fourth piece of the spring 2026 agent puzzle. For the rest, see our hub on the agent layer arriving and the companion pieces on Claude Plan Mode, Claude Design, Notion Custom Agents, and OpenAI AgentKit.