Hub
Framework library
All frameworks and models named by posters, grouped by area.
Every model, method or checklist a poster laid out, merged where different people describe the same thing. None appears in more than one post, so each is listed here with its source.
How Hermes works for a business (6)
- GEPA (genetic-pareto prompt evolution) — An offline optimisation pipeline in a companion repo that reads execution traces, proposes improvements by evolutionary search, scores them with an LLM judge and rubrics, and never commits directly. (Loïc Berthelot · Hermes God-Mode Instruction Manual… ¶ L135)
- Hermes self-monitoring checks — Health checks Hermes designed for itself in a dedicated Telegram chat, with quiet alerts, a daily digest and a weekly audit, mostly as cheap script-only jobs. (an attorney who runs a local business · My Hermes setup (Notion knowledge base)… ¶ L104)
- Hermes three-tier memory — Three memory layers for different purposes: a small always-in-context snapshot, searchable session history, and pluggable external providers. (Loïc Berthelot · Hermes God-Mode Instruction Manual… ¶ L81)
- Notion Kanban project tracker for Hermes — A Kanban board Hermes updates twice daily, with columns for ideas, in progress, active crons, errored crons, done and cancelled. (an attorney who runs a local business · My Hermes setup (Notion knowledge base)… ¶ L69)
- Skill progressive disclosure — Three levels that keep token cost low as the agent loads skills: names first, full content when needed, then specific reference files. (Loïc Berthelot · Hermes God-Mode Instruction Manual… ¶ L101)
- Skill self-improvement loop — The agent solves a problem by trial and error, saves the approach as a SKILL.md file and loads it next time, with a curator that marks skills stale or archives them. (Loïc Berthelot · Hermes God-Mode Instruction Manual… ¶ L116)
What businesses use it for (4)
- FounderOS — Seven Hermes assistants, each owning one function of a solo founder’s startup (growth, marketing, content, SEO, social listening, SRE), coordinated by a Chief of Staff assistant that turns data into daily decisions. (Animesh S. · FounderOS: seven Hermes assistants for a… ¶ L2)
- Hermes-fronted support escalation chain — Hermes gathers customer details and starts a per-customer Pi session on a local model; that agent escalates to Hermes, which escalates to humans. (a founder who replaced their helpdesk · We replaced our helpdesk with Hermes… ¶ L6)
- Planned lead-to-follow-up chain and company rhythm — A per-lead chain (research, ticket, drafted email, reminder) plus fortnightly, monthly and quarterly jobs, all built from skills and cron jobs; stated as the author’s intended direction. (author of the Halfbit Studio blog post · Halfbit Studio: How we run our company with… ¶ L157)
- Slack-driven TAM builder — Hermes on a VPS inside Slack, plugged into data and verification tools, loading a GTM skill from a GitHub repo before each job. (@coldemailchris · Our team built an AI agent that builds our… ¶ L2)
Selling Hermes services (2)
- Demo-led client onboarding — A consultant builds a working version of the client’s own process before selling and demonstrates it live against the real process. (an SME automation consultant · Anyone else deploying Hermes into SME… ¶ L9)
- Hub-and-spoke client setup — An on-premises box the installer controls is the hub; client machines connect as spokes over Tailscale, with Claude max subscriptions for main work and the Hermes API for other workflows. (an installer charging a monthly fee · I made €2,700 last month installing Hermes… ¶ L29)
Deployment and cost (2)
- Multi-person Hermes deployment options — The ways witcheer lists for running Hermes for several people. (witcheer · How do people run Hermes for more than one… ¶ L2)
- Role-based capability layer on one gateway — One gateway identifies each person and applies a role limiting what the agent may do for them, with role dashboards read from an Obsidian vault at session start. (Michael Versluis · How do people run Hermes for more than one… ¶ L13)
Safety and control (1)
- Private-to-public Q&A pipeline with human tag gate — Hermes fills a private Q&A database from sent email; the owner reviews and tags rows, and a separate cron moves tagged rows to the public database the chatbot reads. (an attorney who runs a local business · My Hermes setup (Notion knowledge base)… ¶ L80)
Sources
- (01)
- (02)
- (03)
- (04)
- (05)
Our team built an AI agent that builds our entire TAM (Slack)
- (06)
My Hermes setup (Notion knowledge base), crons and use cases: attorney-operator on a Raspberry Pi
- (07)
- (08)
We replaced our helpdesk with Hermes (comment on 'Salesforce to acquire Fin')
- (09)
How do people run Hermes for more than one person? (incl. Versluis RBAC reply)