My Claude Code chief of staff: one VM, eight specialists and one Telegram chat

Description and use

The system runs on a single Azure VM where a chief‑of‑staff agent delegates tasks to eight specialist agents via a Telegram bot, enabling efficient work without terminal access and reducing token costs.

Problem

Managing diverse tasks without terminal access required a single interface that could delegate work to specialized agents and maintain context efficiently.

Outcome

The system achieves efficient task delegation, reduced token usage, and reliable operation with a single Telegram chat, while keeping specialist context isolated and costs manageable.

Workflow

• 1. User sends text or voice note to Telegram bot. 2. Bot forwards message to chief of staff Claude Code session. 3. Chief of staff reads brief, selects appropriate specialist, and forwards task. 4. Specialist performs work and returns result to chief of staff. 5. Chief of staff relays result back to Telegram chat. 6. Timer loops run scheduled deterministic tasks via headless Claude runs. 7. Memory updates are stored as markdown files and shared across sessions.

Evidence

For the last week, most of my work outside meetings has gone through one conversation. I send a text or a voice note to a Claude Code session on Telegram, and it hands the work to one of eight other Claude sessions, each with its own job.

Source

Everything runs on one Azure VM with a permanently logged‑in desktop, because some work needs a real browser.

Source

The chief of staff never does the work itself. It routes each task to a specialist and relays the result.

Source

The watchdog is a systemd timer that runs a short shell script every two minutes: 30 to 40 milliseconds of CPU and about 5 MB of memory per healthy tick.

Source

On one day early on, the Anthropic dashboard showed about $195, and I did not know why. The audit found that 88% of spend was the model re‑reading its own context.

Source