In one sentence
An open-source, single-tenant personal AI agent from Nous Research that works from a terminal and chat channels.
Hermes Agent is an open-source AI agent built by Nous Research. It runs from a terminal and can also be reached through channels such as Telegram, Slack, Discord and email. It supports skills for repeatable procedures, MCP servers for tools, memory for durable preferences and project facts, and scheduled jobs. Its documentation is at hermes-agent.nousresearch.com and its code on GitHub.
Who it is for
Hermes is designed as a single-tenant personal agent. It suits a founder, operator or small technical team who wants an agent that does real work on their behalf. Shared use should be scoped deliberately rather than treated as a multi-tenant employee platform.
Where it runs
A local workstation, an always-on VPS, the official Docker image or another supported execution environment. The choice affects cost, uptime and, most of all, isolation.
Security is an architecture decision
Hermes includes authorization, approval prompts, credential filtering and container backends. Those controls do not make every deployment safe by default. When an agent reads the open web, inbound email, shared channels, third-party skills or untrusted MCP output, prompt injection is a real risk. Whole-process isolation, where the agent runs inside a container or separate machine so the operating system is the boundary, is safer than sandboxing only the terminal tool. Useful controls include user allowlists or pairing, minimal mounts, scoped tokens, restricted working directories and MCP tool filters.
Memory has limits
Hermes memory is for preferences and facts, not a database. Large or structured data belongs in the systems that already hold it, reached through tools.
Compared with OpenClaw
Both are self-hosted, both run a personal agent reachable from chat. They differ in design, extension model and defaults; our comparison post covers the details.
Our service
We deploy and configure Hermes on infrastructure you control or on an isolated server we operate under a separate agreement. We are independent and not affiliated with Nous Research.
Keep following the system
OpenClaw
An open-source, self-hosted Gateway for a personal AI assistant that connects models, tools, sessions and messaging channels.
AI Agent
A system that uses a model and tools to pursue a goal across multiple steps within defined permissions and review boundaries.
Prompt Injection
An attack where text in an input tricks an AI model into ignoring its instructions or taking actions it should not.
Model Context Protocol (MCP)
An open standard for connecting compatible AI applications to tools, resources, and prompts.
Buzz (Block)
An open-source, self-hostable team workspace from Block where people and AI agents share rooms on a signed Nostr event log.
Subagent
A separate AI agent session that a main agent starts for one task, with its own context window, instructions and tools, which returns a short report when done.
Context Compaction
Replacing the older part of an AI agent's conversation with a written summary when its context window nears full, so a long session can keep going.