Hermes Agent Demo: Local AI Beats OpenClaw

Installation and Onboarding

The video shows installation via a single command: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash. Users then run hermes setup to add API keys and enable chat interfaces in terminal, Slack, or Telegram. Hermes Agent runs locally on laptops, stores data locally, and uses user-provided API keys.

Dashboard and Built-in Capabilities

Hermes Agent includes a dashboard with sessions view (recent chats, cron jobs, connected platforms like Slack), analytics (token usage, session count, model usage, top skills), models list, logs for auditing actions, cron jobs management, and a skills page listing 87 built-in skills filterable by category. Examples include Apple tools (Find My, iMessage, Apple Notes) and media tools (Spotify, YouTube). It supports over 80 integrations at launch, such as Spotify, Slack, GitHub, Notion, and Linear. Users manage plugins by installing from GitHub, customize dashboard colors (e.g., Hermes teal), and view stored API keys/secrets. Capabilities cover coding, shell commands, file management, Spotify control, research, automation (cron jobs, webhooks, websites, PDFs), web search, browser control, project sandboxes, and parallel agent delegation. The agent autogenerates skills from conversations and captures memory.

Multi-Agent Profiles

Profiles assign models per role, editable via dashboard. Default uses Kimmy K2.6. Sub-agents include Designer (Opus 4.7; SOUL.md prompt: "your visual communication specialist your job is to create compelling visuals slides HTML artifacts things like that" with rules/preferences), Researcher, Reviewer, Runner (Haiku model for cost efficiency), and X Writer ("you're the social voice of this weekend AI you write ex and Twitter posts that promote podcast episodes while delivering standalone value", including voice, formats, examples). Users edit SOUL.md system prompts directly and set default models/context window sizes.

Kanban Board Automation

The Kanban board tracks tasks in columns: triage (raw ideas), to-do, waiting (dependencies), ready, in progress, blocked, review, done. A cron job runs every 10 minutes to audit the board, triage tasks, enrich descriptions (e.g., adding context, deliverables, competitors like for "research competitors for this weekend AI newsletter", analysis methods), and auto-move tasks (e.g., triage to ready). Users assign ready tasks without manual column shifts. The video contrasts this with OpenClaw's weaker dashboard, claiming Hermes excels in memory, automations, and usability.