A web dashboard that actually runs your AI agents, not just chats with them
Hermes Workspace is a React control center for the Hermes Agent framework, bundling chat, terminal, memory, skills, and multi-agent swarm orchestration into one install.

What it does Hermes Workspace is a React/TypeScript web UI that wraps the Hermes Agent ecosystem into a single pane. You get chat with SSE streaming, a Monaco file browser, a cross-platform PTY terminal, memory search and editing, a skills marketplace, and a dashboard tracking sessions, model mix, and costs. It installs via one shell script and connects to either a local Hermes Agent gateway or any OpenAI-compatible backend.
The interesting bit
The “zero-fork” architecture is the careful part: the workspace attaches to vanilla NousResearch/hermes-agent through official installers, with no source patches. If the dashboard API is missing, Conductor falls back to a native Swarm dispatch mode rather than failing. Swarm Mode goes further, turning the UI into a live control plane with tmux-backed persistent workers, role-based routing, and a byte-verified review gate before PRs ship — all surfaced through a built-in TUI view.
Key highlights
- Swarm Mode: multi-agent orchestration with persistent tmux workers, Kanban task board, and proof-bearing checkpoints
- Capability gates: features needing upstream endpoints show clean placeholders instead of mid-action crashes
- PWA + Tailscale support for phone-to-homelab access without tunnel gymnastics
- 2,000+ skills catalog with origin badges, filters, and source paths
- Auth middleware on every route, CSP, path-traversal guard, fail-closed remote bind
- Works with local models (Ollama, LM Studio, vLLM, Atomic Chat) in portable mode without the full agent gateway
Caveats
- Full feature unlock requires two separate services: gateway (
:8642) and dashboard (:9119); misconfigure one and enhanced panes stay locked - Node 22+ required; the “~2 min” Docker quickstart claim is in the docs but not independently verified here
- Swarm Mode’s autonomous PR/issue lanes and “repair playbook” are described aspirationally — actual autonomy level depends on upstream agent behavior
Verdict Worth a look if you’re already running Hermes Agent or want a self-hosted alternative to closed AI dashboards. Skip it if you need a drop-in chat wrapper without the infrastructure appetite — this wants to be your agent’s operating system, not its window.