Hermes Agent Dashboard and Kanban: 7 Features That Now Beat OpenClaw

Key Takeaways

• Start the dashboard with hermes dashboard — it opens a local URL with full task, Kanban, memory, and agent management UI.
• Kanban board columns: triage → todo → ready → in progress → blocked → done. Tasks flow through these columns as agent workers pick them up and execute.
• Unlike chat (single thread, tasks queue sequentially), the Kanban board allows 10-30 simultaneous tasks with dedicated worker agents handling each one in parallel.
• Recommended setup: a "librarian" agent running on ChatGPT (cheap) checks the board every 10 minutes, fleshes out triage tasks using stored memories, and moves them to ready. A separate main worker agent handles execution.
• Librarian cron prompt: "Check Kanban every 10 minutes. If tasks are in triage, flesh out details from memories and move to ready. If tasks in ready have an agent assigned, execute them."
• OpenClaw's two problems: (1) every update breaks the tool — the reviewer stopped updating because the repair time is too high; (2) performance bloat from too many unrelated features shipped at once.
• Hermes updates have themes — all features in a single release tie together into a coherent narrative, making updates predictable and safe to apply.

The Kanban Board: Why It Changes Multi-Tasking

When you talk to Hermes through Telegram, iMessage, or Discord, you're operating in a single thread. You send a message, the agent works on it, and subsequent messages queue up. If you're managing 20 different projects — building a SaaS, writing content, managing a community, handling research — this single-thread model is a bottleneck. The Kanban board breaks that constraint.

Alex demonstrates adding three tasks to triage at the start of his workday. Within 60 seconds, his librarian agent (running on a cheap model via a cron job) reads the board, looks through his stored Obsidian memories for relevant context, fills in the task details, and moves each task to "ready." He then assigns each task to a worker agent, which picks it up and moves it to "in progress" automatically. The task shows agent comments as it progresses, and lands in "done" when complete. The only steps he takes manually: writing the triage task names and assigning the agent when tasks reach "ready."

The Two-Agent Architecture

Running the Kanban manager cron on your main agent creates a conflict — the manager fires every 10 minutes and can block your main agent from doing real work. The solution is a dedicated second agent (the "librarian") that handles all administrative work: board management, task fleshing, scheduling, and low-stakes operations. The main worker agent stays focused on actual execution.

To set this up: tell your main Hermes agent "Set up another Hermes profile for a librarian." It creates the profile automatically. Then instruct the librarian to run the Kanban check prompt every 10 minutes. The librarian can run on a cheap model (Alex uses ChatGPT) since its work is mostly context retrieval and board manipulation, not complex reasoning. This two-agent pattern scales — you can add specialized worker agents for specific task types (coding, content, research) and the librarian routes work to the right one.

Commands & Code Mentioned

hermes dashboard

Related on OpenClawDatabase

• Hermes Agent Setup Guide
• Hermes Tasks — cron jobs and scheduled work
• Hermes vs OpenClaw — detailed platform comparison
• Hermes Memory — skill files and memory storage