Agent stack build plans · № 2

Hermes as an omnichannel community manager

One always-on agent in your Discord, Slack, and WhatsApp — answering with real data, keeping notes on members, and queueing the weekly shoutouts.

CLI: hermes + gateway daemon MIT · 219k GitHub stars v0.19.0 “Quicksilver” · July 2026 model-agnostic · bring your own keys

Hermes is Nous Research’s open-source agent — a terminal CLI plus a messaging gateway daemon that puts the same agent, same memory, same skills into Discord, Slack, and WhatsApp at once (and about twenty other platforms, if you must). Group chats are supported. That shape is exactly a community manager’s job: be everywhere the community is, remember everything, and never sleep.

The honest version: Hermes gives you the gateway, the memory files, the scheduler, and — the good part — a skill_manage tool that lets the agent write its own skills to pull external data. It does not ship per-member memory or automated rewards; this plan has you build both as skills, and marks every such seam “you build this”. Effort: an evening to get the gateway talking on all three platforms, a weekend including the notes and rewards glue. Cost: an always-on agent pays a model call for every message it handles — put a cheap model behind it (OpenRouter or your own endpoint; Hermes doesn’t care) — and Nous Portal’s web tools are metered extras, not requirements.

The plan, in five parts

1What you getthe community-pulse dashboard mock, and a day in the life of the bot 2Architecturegateway, markdown memory, skills, cron — and the three parts you build 3Set it upinstall, connect three platforms, lock the door — real commands 4The promptsstaged, paste-able — brief it, make it write its own skills, wire the rewards cron 5The trapsallow-all auth, self-editing skills, and the 2,200-character memory

Is this for you?

Yes, if you run a community that’s outgrown one platform — an OSS project with a Discord, a Slack for partners, a WhatsApp group for the core — and the same questions land in all three. Yes, if you want the answering bot and the housekeeping bot to be one brain. Skip it if your community lives on a single platform (a native bot framework is less machinery) or if you need tiered moderator permissions inside one bot — Hermes’s access control is deliberately binary, and this plan works with that rather than around it. We also don’t cover moderation (auto-deleting, banning): wiring an LLM’s judgement to a ban hammer is a different plan and a worse idea.