AI Agentic Operations — Telegram & VPS

A real Telegram-connected assistant that helps manage development work and routine operations on this VPS, with human approval for high-risk actions. The system is built around Hermes Agent, Codex CLI for substantial coding work, Docker Compose for application runtime and bounded host operations through agentctl.

Routine deployment, troubleshooting and health checks often require manual SSH sessions and fragmented context. The goal is to make normal workflows available in a chat while preserving clear security boundaries.

A Telegram interface delegates code work to Codex, inspects project files and Git state, runs tests and health checks, and maintains project notes. Risky changes are surfaced for explicit owner approval; the web applications do not gain arbitrary host shell access.

Telegram → Hermes Agent → specialized development/diagnostics tools → project containers and bounded agentctl host operations. PostgreSQL backs managed applications; HTTPS ingress serves the public portfolio. The public case study omits internal addresses, bot credentials and private operational details.

Hermes Agent, Codex CLI, Telegram, Docker Compose, PostgreSQL, Next.js, TypeScript, Caddy and agentctl.

Balancing convenience with least privilege: keeping credentials outside the portfolio app, retaining human approval for destructive actions, verifying changes with real test and runtime output, and documenting state across sessions.

The assistant operates through this Telegram conversation and has been used to build and deploy the portfolio CMS and run verified server health checks. It is an ongoing operational system, not an autonomous no-approval production administrator.