Hermes WebUI is the MIT-licensed browser frontend for Hermes Agent, an always-on autonomous agent that runs on a user-controlled server. With 11,400+ GitHub stars and 1,500+ forks, it has become one of the most-starred self-hosted agent UIs of 2026, offering a dark, mobile-responsive interface that replaces the usual terminal-only access to long-running agent systems. ## What Hermes WebUI Solves Most open agent frameworks ship either a CLI or a developer-facing TUI. That is fine for engineers but a poor match for an agent that is supposed to run 24/7 on a server, schedule its own tasks, and act on behalf of the user across messaging platforms. Hermes WebUI fills that gap with a real web app: log in from a phone or laptop, see all sessions, watch the agent stream replies in real time, and manage its files without SSHing in. ## Three-Panel Layout The core interface is split into a session sidebar, a chat pane, and a workspace file browser. The sidebar tracks every conversation the agent is participating in, including ones triggered by external messaging platforms. The chat pane streams responses via Server-Sent Events so output appears token-by-token without manual refreshes. The file browser exposes the agent's working directory directly, which matters because Hermes is allowed to read, write, and edit files as part of its task execution. ## Multi-Provider Backend Hermes Agent itself is provider-agnostic. The WebUI exposes that flexibility, supporting OpenAI, Anthropic, Google, DeepSeek, and several other providers from a single configuration surface. Users can switch the model behind a running session without rewriting their workflows, which is important for cost tuning (a cheap model for routine triage, a frontier model for harder reasoning). ## Persistent Memory and Self-Improving Skills Unlike chat UIs that treat each session as ephemeral, Hermes WebUI surfaces the agent's persistent memory layer. The agent retains information across sessions, learns new skills from repeated interactions, and can be queried about what it remembers. This is the feature that makes the self-hosted angle meaningful: a server-resident agent with continuous memory becomes more useful over time, while a stateless cloud chatbot does not. ## Messaging Platform Integrations Hermes ships connectors for more than ten messaging platforms, including Telegram, Discord, and Slack. The WebUI is where users wire those connectors up, see incoming messages from them, and audit what the agent did in response. Combined with self-hosted cron-style scheduling, this turns Hermes into a personal automation server that can run unattended jobs and report results through the user's preferred chat platform. ## Voice, Themes, and Accessibility Voice input is wired through the Web Speech API, and the UI ships with multiple themes and skins. Mobile responsiveness is treated as a first-class concern, with a hamburger navigation pattern designed for phone screens. These are not headline features in isolation, but together they make the difference between a project that sits on a server and a project that someone actually uses from their pocket. ## Authentication Options Because an autonomous agent on a public server is a serious attack surface, Hermes WebUI supports both classic password login and WebAuthn passkeys. The WebAuthn path is recommended for production deployments since it eliminates the phishing surface that a long-lived agent endpoint would otherwise expose. ## Limitations The project is Python plus vanilla JavaScript, which keeps the deployment story simple but means contributors have to work without a modern frontend toolchain. The agent's autonomy is only as safe as the providers and tools it is wired to, so giving Hermes access to filesystem operations, messaging platforms, and external APIs is a security decision that the user owns end to end. Self-hosting still requires a real server, a domain, and TLS, which raises the barrier for non-technical users despite the polished web UI. And because Hermes Agent itself is a separate dependency, the WebUI is only as capable as that backend at any given release.