Hermes Workspace is a web UI for the Hermes Agent runtime from Nous Research. It bundles chat, sessions, agent memory, skills, MCP servers, scheduled jobs, a file browser with a Monaco editor, and a terminal into one browser interface, so the agent is operated from a single place rather than from the CLI.
The workspace runs as a Node.js 22+ app on port 3000 and talks to two services: the Hermes gateway on port 8642 for chat, models, streaming, and jobs, and the Hermes dashboard on port 9119 for sessions, skills, config, and MCP. It follows a zero-fork model: it runs against an unmodified hermes-agent install, and features that need endpoints the backend lacks show a placeholder instead of failing.
It also works in a portable mode against any OpenAI-compatible endpoint, such as Ollama, LM Studio, vLLM, or Atomic Chat, where chat works but sessions, memory, and skills report as not available. The full feature set unlocks when both Hermes services are reachable. A Swarm mode turns the workspace into a control plane for tmux-backed Hermes worker agents with role-based dispatch, a Kanban task board, and a reports inbox.
Features
- Chat: SSE streaming with tool-call rendering, multi-session support, and markdown with syntax highlighting
- Memory: browse, search, and edit agent memory in a live markdown editor
- Skills and MCP: browse skills with origin badges and filters, plus an MCP page with catalog, marketplace, and sources
- Files and terminal: workspace file browser with Monaco and a cross-platform PTY terminal
- Operations: multi-agent dashboard with profile presets (Sage, Trader, Builder, Scribe, Ops) and setup detection
- Conductor: mission dispatch and decomposition, using the dashboard mission API when present and native Swarm dispatch otherwise
- Swarm mode: persistent tmux-backed worker agents, role-based routing, a task board, and a built-in TUI view for attaching to workers
- Dashboard: sessions, model mix, cost ledger, and an attention card in one overview
- Backends: Anthropic, OpenAI, OpenRouter, Ollama, vLLM, LM Studio, Atomic Chat, or any OpenAI-compatible chat-completions endpoint
- Themes and PWA: Hermes, Nous, Bronze, Slate, and Mono themes in light and dark, installable as a PWA and reachable over Tailscale
- Security: auth middleware on every route, CSP, a path-traversal guard, and fail-closed remote binding
