Hermes-Agent's value in 10% the code,
by riding Claude Code's native rails.
A messaging gateway, voice in/out, persistent project-leads with their own agent teams, all reachable from a phone — built as a Claude Code plugin instead of a 27,000-LOC platform.
Architecture
One always-on machine. One persistent Claude session. A handful of MCP servers and skills. Everything else — channels, cron, teams, memory, mobile — is native to Claude Code.
Telegram ──► claude --channels (OCI VPS, Max OAuth)
│
├── voice_stt MCP ──► whisper.cpp
├── voice_tts MCP ──► piper
├── project_orchestrator MCP ──► spawns project-leads
└── hermes_api MCP ──► FastAPI ──► claude.mayankgupta.in
Each project-lead is its own background claude session with its own
TeamCreate-instantiated team. The orchestrator never team-leads itself.The thesis
Hermes-Agent by Nous Research is a remarkable platform for self-improving messaging agents — 27,000 lines of Python implementing channels, cron, skill systems, memory curation, sandbox backends, and trajectory tooling for model training.
After studying it, I asked: what if the engine is Claude Code? Claude Code already has channels (Telegram, Discord, iMessage, custom), agent teams, server-hosted scheduled routines, agent view, Remote Control, mobile push, MCP, hooks, plugins, and auto-memory. The platform layer collapses.
Hermes-Claude is the answer: a Claude Code plugin (~4,000 LOC) that fills the last 5% — a voice pipeline, a project orchestrator, a dashboard, and curated workflows. No API keys, just Claude Max subscription. Runs on a single Oracle Cloud free-tier VPS. The 90% you're looking at on this page isClaude Code, not me.
Trading integrations come in V2; this V1 is the platform demonstration.
See it run
A 60-second walkthrough is in progress for V1.5. In the meantime, the dashboard is live — login is restricted, but the public stats above pull from the running system.