AI Infrastructure

Hermes Dashboard

Definition
The Hermes Dashboard is the browser-based control plane of the open-source Hermes agent by Nous Research, launched with the command `hermes dashboard` on localhost port 9119. Where a classic agent CLI ends at the terminal prompt, the dashboard turns the running agent into an operable system: the Status page shows gateway state, connected platforms, and the 20 most recent sessions with model, message count, and token usage, auto-refreshing every 5 seconds. The Chat tab embeds the real TUI in the browser via xterm.js over a WebSocket-backed pseudo-terminal, so slash commands and approval prompts behave identically to the terminal. The Config editor exposes 150+ configuration fields as forms, the API Keys manager replaces .env hunting, the Sessions browser makes history full-text searchable via FTS5, and the Cron page lists every scheduled job with last-run output and next trigger — a failed 2 a.m. automation is diagnosable without SSH or grep. Technically it is a FastAPI backend with a Vite/React/TypeScript/Tailwind frontend, installed as an optional extra via `pip install 'hermes-agent[web,pty]'`. Since release v2026.9.24 (September 24, 2026), binding to a non-loopback address requires prior registration (`hermes dashboard register`, Nous Portal OAuth); Docker deployments set HERMES_DASHBOARD=1. The distinction to an agent harness: the harness is the execution loop wrapped around the model — tools, sandbox, retries — while the dashboard is the operator layer above it. Against hosted control planes such as the Claude Code desktop app or Codex's workspace UI, everything stays self-hosted: logs, session history, and credentials remain on your own server, a posture the project backs with roughly 248,000 GitHub stars (September 2026).
Category
AI Infrastructure

Deep Dive: Hermes Dashboard

The Hermes Dashboard is the browser-based control plane of the open-source Hermes agent by Nous Research, launched with the command `hermes dashboard` on localhost port 9119. Where a classic agent CLI ends at the terminal prompt, the dashboard turns the running agent into an operable system: the Status page shows gateway state, connected platforms, and the 20 most recent sessions with model, message count, and token usage, auto-refreshing every 5 seconds. The Chat tab embeds the real TUI in the browser via xterm.js over a WebSocket-backed pseudo-terminal, so slash commands and approval prompts behave identically to the terminal. The Config editor exposes 150+ configuration fields as forms, the API Keys manager replaces .env hunting, the Sessions browser makes history full-text searchable via FTS5, and the Cron page lists every scheduled job with last-run output and next trigger — a failed 2 a.m. automation is diagnosable without SSH or grep. Technically it is a FastAPI backend with a Vite/React/TypeScript/Tailwind frontend, installed as an optional extra via `pip install 'hermes-agent[web,pty]'`. Since release v2026.9.24 (September 24, 2026), binding to a non-loopback address requires prior registration (`hermes dashboard register`, Nous Portal OAuth); Docker deployments set HERMES_DASHBOARD=1. The distinction to an agent harness: the harness is the execution loop wrapped around the model — tools, sandbox, retries — while the dashboard is the operator layer above it. Against hosted control planes such as the Claude Code desktop app or Codex's workspace UI, everything stays self-hosted: logs, session history, and credentials remain on your own server, a posture the project backs with roughly 248,000 GitHub stars (September 2026).

Tech Stack
Nous ResearchFastAPIReactTypeScript

Production-Ready Guardrails