Case study · #01 / AI agent

Glyph Personal Agent

A personal AI agent wired into daily life — one Telegram thread, several modes, running itself on cron. Not a demo; my actual workflow.

Nightly ingest 12+ integrations Runs on cron
Role
Dev + design
Surface
Telegram
Integrations
12+
Runtime
24/7 · cron
Status
Live · daily
The idea

I wanted an assistant that lived inside my day, not a chatbot I'd open and forget. One surface I already use — Telegram — that could brief my calendar, track a job search, log how I'm doing, run code, and reach out on its own. Built for exactly one user: me. Not a demo, my actual workflow.

The architecture

Messages hit a Hermes gateway that routes them to the right mode — Glyph for general coordination, Solace for wellbeing, Neo for technical work, Spark for creative tasks, Boss for career strategy. Same Claude reasoning core underneath; each mode shifts its register, tone, and tool set. It runs as an MCP setup on a persistent Ubuntu server, wired to Google Calendar, GitHub, and Drive, with Honcho holding cross-session memory so facts and project state persist without being re-stated.

How it runs

Eight cron jobs deliver proactive updates on a schedule — daily briefings, check-ins, reports — without me prompting. Twelve-plus integrations, thirty-plus tools, one chat thread. It has run every day for roughly half a year: the morning briefing, the wellbeing log, the code execution, the career nudges, all arriving on their own.

For1 user (me)
SurfaceTelegram
Modes5
Integrations12+
Cron jobs8
Days running180+
Reasoning coreClaude
Want one like this?
Schedule a call →