A unified workspace and orchestration command center for AI Agents — integrating chat, file browsing, memory & skill management, terminal, multi-agent collaboration, and Swarm mode, compatible with any OpenAI-compatible backend.
Hermes Workspace is a unified workspace for AI Agents, positioned as a "command center" for agent interactions. It consolidates chat, file browsing & editing (Monaco Editor), memory management, a skill marketplace (2,000+ skills), MCP tool directory, cross-platform terminal (xterm.js), multi-agent orchestration dashboard, Swarm mode (tmux-based persistent workers with Kanban boards), and an ops dashboard (session stats, cost ledger, model mix) into a single interface.
The project adopts a zero-fork architecture, running directly on the upstream NousResearch/hermes-agent without code modifications. It uses a dual-service architecture (Gateway :8642 + Dashboard :9119) to connect to any OpenAI-compatible backend (Anthropic / OpenAI / OpenRouter / Google / Ollama / LM Studio / vLLM, etc.). The frontend is built on React 19 + Vite 7 + TanStack Router + Zustand 5 + Tailwind CSS 4, with support for SSR, PWA installation, Electron desktop packaging, and Tailscale remote access.
Swarm Mode supports role-based task dispatch (builder / reviewer / docs / research / ops / triage / QA / lab) with YAML configuration, Kanban boards, checkpoint reports, and byte-verified review gates for automated pipelines. Security covers per-route Auth middleware, CSP, path traversal protection, and fail-closed remote binding.
Deployment options include a one-click script (macOS/Linux), Docker Compose, or manual connection to an existing hermes-agent instance, with portable mode (chat only) and enhanced mode (full features) available.
Known limitations: Conductor (task dispatch & decomposition) depends on an upstream dashboard plugin and is currently a placeholder (Issue #262); Swarm Mode depends on tmux and is limited on non-Unix environments; no one-click install script for Windows yet (Docker or WSL required).