AI harnesses are the software layer that makes a language model capable of acting. Hermes Agent is Nous Research’s open-source AI harness: a self-improving AI agent with persistent memory that runs in the terminal, as a desktop app, in IDEs and through messaging platforms, and works with nearly any language model. As of 20 September 2026.
What is Hermes Agent?
Hermes Agent is developed by Nous Research, a research lab that also publishes the Hermes model family. The project is open source on GitHub at NousResearch/hermes-agent under the MIT licence. On Linux, macOS, WSL2 and Termux it is installed via curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash; native Windows uses a PowerShell one-liner. Afterwards, hermes starts an interactive conversation. Besides the terminal interface, Hermes Agent offers a desktop app, an ACP integration for VS Code, Zed and JetBrains, and a gateway that, per the documentation, connects to 20+ messaging and chat platforms, including Telegram, Discord, Slack, WhatsApp, Signal and Microsoft Teams.
Hermes Agent as an AI harness
Technically, a synchronous agent loop (the AIAgent class) drives the cycle of provider calls, tool execution and context compression - the same class serves the CLI, gateway, ACP integration and batch processing, per the architecture documentation. Tools range from file access to web search and browser automation, and can be extended with any MCP server. What stands out is the skills system: for recurring tasks, the agent automatically writes SKILL.md files into ~/.hermes/skills/, improves them with use, and loads their full content only when needed - compatible with the open agentskills.io standard. Risky commands are governed by a tiered approval model (Smart, Manual, Off) with a hard-coded blocklist for irreversible actions such as rm -rf /. For parallel workstreams, the agent can spawn isolated subagents via tools/delegate_tool.py. Per the documentation, a Python plugin system covers custom tools, providers, memory backends and read-only observer hooks for telemetry; for external programs, the Agent Client Protocol (ACP), a JSON-RPC gateway protocol and an OpenAI-compatible HTTP API (gateway/platforms/api_server.py) are additionally available for programmable integration.
Models and model access
Hermes Agent is deliberately model-agnostic: hermes model switches between Nous Portal, OpenRouter, Anthropic, OpenAI, Google Gemini, AWS Bedrock, Azure AI Foundry and self-hosted options such as Ollama or vLLM. For a custom endpoint, set provider: custom and base_url under model: in ~/.hermes/config.yaml, or choose “Custom endpoint” interactively in hermes model. Because this freely configurable OpenAI-compatible endpoint is documented, model access can generally be routed through a central proxy such as our AI Gateway by entering its endpoint as base_url.
Enterprise use: data protection and governance
The security documentation describes a layered model of approval modes, blocklists, container isolation for the Docker backend, and SSRF protection for web tools. It does not address telemetry or central usage-data collection, and no enterprise SSO is documented; instead, DM pairing codes and allowlists govern access via messaging channels. Importantly for practice: source code, file contents and prompts go to whichever model provider is configured - with cloud providers, that means outside your own environment. This belongs in your records of processing activities in advance, regardless of whether Hermes Agent runs locally or on a server.
Orchestrating Hermes Agent with herdr
herdr is a terminal runtime that supervises several coding agents at once. Per herdr’s documentation, it detects Hermes Agent through the terminal screen manifest and provides session integration through it - deeper state integration via lifecycle hooks is not currently documented for Hermes Agent. Teams running several terminal agents in parallel can fold Hermes Agent into the same overview as other supported harnesses this way.
Where it fits: developer harness and enterprise harness
Hermes Agent is primarily a personal, self-improving agent for developers and technically inclined users - with a terminal, IDE integration and its own messaging channels. For business users across an entire company who don’t work from a terminal, CompanyGPT is the fitting harness: tools via MCP servers and integrations, roles from Entra ID or Keycloak, and audit logs for governance. Our AI Gateway provides the shared cost and governance layer across different harnesses and model providers. For an overview of other agent harnesses, see our harness overview. If you’d like to check how a self-improving agent like Hermes Agent fits securely into your existing system landscape, we’re glad to help with selection and hardening.