Hermes WebUI is a self-hosted browser front end for the Hermes agent. It aims at near 1:1 parity with the Hermes CLI, so everything available from the terminal is available in the browser, and it uses the existing Hermes configuration, skills, memory, and model providers without a separate setup. The server is a Python standard-library HTTP server with a vanilla JavaScript front end: no build step, framework, or bundler.
The layout has three panels: a left sidebar for sessions and navigation, chat in the center, and a workspace file browser on the right. Model, profile, and workspace controls sit in the composer footer, a circular context ring shows token usage, and settings and session tools live in the Hermes Control Center. By default the WebUI runs the Hermes agent in-process from the agent's own configuration. Optional modes route browser chat through a running Hermes Gateway or attach session-recall prefill from Joplin, Obsidian, Notion, or llm-wiki routers.
Chat streams tokens over server-sent events with automatic reconnection across SSH tunnel drops, shows tool calls and subagent delegation as inline cards, renders Mermaid diagrams and extended-thinking blocks, and asks for approval before dangerous shell commands. Sessions can be pinned, archived, grouped into colored projects, tagged, searched, exported as Markdown or JSON, or shared as a sanitized read-only link, and CLI sessions from the agent's SQLite store appear in the same sidebar for import. The workspace browser previews and edits text, code, Markdown, and images, handles file and folder operations, and shows the Git branch and dirty-file count.
Models come from any Hermes API provider (OpenAI, Anthropic, Google, DeepSeek, Nous Portal, OpenRouter, MiniMax, Xiaomi MiMo, Z.AI), with a dropdown built from the configured keys and a /model command that resolves the agent's model aliases server-side so the browser never receives a base URL or key. Profiles switch without a server restart and can be created with a custom endpoint for Ollama, LM Studio, or other local servers. Authentication is off for same-machine use and can be enabled as a password, passkeys (WebAuthn), or native OIDC login, with signed HTTP-only cookies, security headers, and SRI-pinned CDN resources. It runs on Linux, macOS, or Windows via WSL2, as a Nix flake or NixOS module, or as single- or multi-container Docker deployments, and native clients for web, Windows, Linux, macOS, iOS, and Android connect to the one server. WebUI and Hermes agent releases are tested together and are meant to be upgraded as a pair.
Features
- CLI parity: chat, sessions, cron tasks, skills, memory, profiles, todos, and workspaces from the browser, using the existing Hermes agent setup and models
- Three-panel layout: sessions sidebar, chat, and a workspace file browser, with model, profile, and workspace controls in the composer footer and a token-usage context ring
- Streaming chat: SSE token streaming with auto-reconnect, inline edit and regenerate, retry, cancel, message queuing, tool call and subagent cards, Mermaid diagrams, and approval cards for dangerous shell commands
- Sessions: pin, archive, projects, tags, search, Markdown and JSON export or import, public read-only share links, and a bridge that imports CLI sessions from the agent's SQLite store
- Workspace browser: directory tree, breadcrumbs, inline preview of text, code, Markdown, and images, file and folder editing, binary download, and Git branch and dirty-count badge
- Multi-provider models: OpenAI, Anthropic, Google, DeepSeek, Nous Portal, OpenRouter, MiniMax, Xiaomi MiMo, and Z.AI, with a model dropdown from configured keys and server-side alias resolution
- Profiles: create, switch, clone, and delete agent profiles without a restart, with optional custom Base URL and key fields for Ollama, LM Studio, and other local endpoints
- Authentication: optional password, passkeys (WebAuthn), or native OIDC login, signed HTTP-only cookies, security headers, a POST size limit, and SRI-pinned CDN assets
- Control Center panels: Tasks (cron jobs with run history and completion alerts), Skills, Memory (MEMORY.md and USER.md), Profiles, Todos, and Spaces
- Slash commands: autocomplete for built-ins such as help, clear, compress, model, workspace, new, usage, and theme, with skills matched by keyword and unknown commands passed to the agent
- Themes and mobile: system, dark, and light themes with eleven skins, voice input via the Web Speech API, and a responsive layout with a slide-in sidebar and 44px touch targets
- Deployment: Linux, macOS, or WSL2, a Nix flake and NixOS module, single- and multi-container Docker, SSH tunnel or Tailscale remote access, and native clients for every device connecting to one server
