The idea
I wanted a single AI system that could actually operate — not just chat. Hermes is a personal autonomous-agent platform I built on top of an open-source agent runtime, where I authored the profiles, the 100+ reusable skills, and the analysis frameworks that give it real capabilities.
Honesty note: Hermes runs on an open-source agent runtime (Nous Research). My work is the system design on top: the multi-profile architecture, the skills library, the finance/career frameworks, and the orchestration & scheduling that make it useful.
How it works
- Multi-profile architecture — isolated “personas” (investing, finance, career) each with their own config, memory, skills and scheduled jobs, sharing one runtime.
- Skills library — 100+ composable, self-documenting skills spanning research, finance, dev-ops, ML and creative work; the agent picks and chains them per task.
- Persistent memory — durable, searchable context so the agent remembers decisions, watchlists and preferences across sessions.
- Tool orchestration — function-calling across web search, files, APIs and CLIs, with multi-agent delegation for bigger jobs.
- Multi-channel & cron — reachable over Telegram, Discord, Slack and email, with scheduled routines that run autonomously and report back.
Why it matters
Hermes is where my product instincts and engineering meet: scoping ambiguous goals into composable capabilities, designing for isolation and reliability, and shipping a system that does real work unattended — the same muscles I use leading enterprise products.