Hermes Workspace is a self-hosted web UI and control surface for the Hermes Agent (NousResearch/hermes-agent). It bundles chat with SSE streaming and tool-call rendering, a memory browser/editor, a skills browser, an MCP catalog, a file browser with a Monaco editor, and a cross-platform PTY terminal into one interface. It runs against vanilla upstream hermes-agent ("zero-fork") and degrades to a "portable mode" of basic chat against any OpenAI-compatible backend when the extended gateway/dashboard APIs are unavailable.
Its multi-agent side ("Swarm Mode" and "Conductor") turns the workspace into an orchestration plane: persistent tmux-backed agent workers receive role-based dispatch (builder, reviewer, docs, research, ops, QA, and similar lanes), a Kanban task board tracks work through backlog/running/review/done states, and a reports inbox surfaces checkpoints, blockers, and handoffs. Conductor decomposes and dispatches missions via the dashboard mission API when present, falling back to a native swarm dispatch otherwise.
It connects to the agent over two HTTP services — a gateway (chat, models, streaming, jobs) and a dashboard (sessions, skills, config, MCP) — and works with hosted and local model providers including OpenAI-compatible endpoints, OpenRouter, Google, Ollama, LM Studio, vLLM, and Atomic Chat. It ships as a PWA installable on desktop and mobile, supports remote access over Tailscale/LAN, and includes auth middleware, path-traversal guards, and a fail-closed guard that refuses non-loopback binding without a password.
