daily-llm-usage-report

name	daily-llm-usage-report
description	Gordon's daily LLM usage report — request metrics sent to Telegram each morning at 6 AM PT. Uses a daemon + cron, no OpenRouter dependency.
version	1.0.0
author	Hermes Agent
license	MIT
metadata	{"hermes":{"tags":["telegram","usage","metrics","cron","daily-routine"],"homepage":"https://github.com/NousResearch/hermes-agent"}}

daily-llm-usage-report

Gordon's daily LLM usage report — sends request metrics to Telegram at 6 AM PT every morning.

Metrics Reported

• User Interactions (24h)
• API Requests (24h)
• API Requests by Model (24h) — model names shortened (e.g. openai-codex/gpt-5.5 → gpt-5.5), sorted descending by call count
• Max Req / 5H Window (24h)
• Max Req / 5H Window (30d)

Components

/opt/hermes/scripts/cost_report.py

Core report script. Reads request log, computes metrics, sends Telegram message. No OpenRouter dependency — Gordon is on subscription.

Usage:

# Test/send report
python3 /opt/hermes/scripts/cost_report.py

# Take baseline (no send)
python3 /opt/hermes/scripts/cost_report.py --baseline

/opt/hermes/scripts/cost_daemon_wrapper.py

Daemon wrapper that reads Telegram tokens from /opt/data/.env.tokens and execs the daemon with those tokens injected into the environment.

Always start the daemon via this wrapper, not directly. The daemon reads tokens from its process env, and cron/the subprocess don't have the container's env vars.

/opt/hermes/scripts/cost_report_daemon.py

Sleeps until next 6 AM PT, runs cost_report.py, repeats. Daemon is long-lived.

/opt/data/request-log.jsonl

JSONL log of API calls. Each line: {"ts": unix_timestamp, "api_calls": int, "platform": "telegram", ...}

/opt/data/cost-state.json

State file (keeps prior metrics for comparison).

Token Setup

Tokens live in /opt/data/.env.tokens:

TELEGRAM_BOT_TOKEN=<bot_token>
TELEGRAM_HOME_CHANNEL=<chat_id>

Starting / Restarting the Daemon

Always use the wrapper to get tokens:

python3 /opt/hermes/scripts/cost_daemon_wrapper.py &

To verify it's running:

ps aux | grep cost_report_daemon
tail /opt/data/cost-report-daemon.log

Pitfalls

The daemon must be started via cost_daemon_wrapper.py (not directly), which reads tokens from /opt/data/.env.tokens. Starting it directly causes silent failures — the daemon sleeps and never sends.

Old daemon still running with stale tokens

If you update /opt/data/.env.tokens, kill and restart the daemon:

pkill -f cost_report_daemon
python3 /opt/hermes/scripts/cost_daemon_wrapper.py &

Cron schedule is UTC

Hermes cron uses UTC. To fire at 6 AM PT, schedule as 5 13 * * * (13:05 UTC). Always verify UTC equivalent.

Request log may be empty on first days

Metrics will be 0 until the log has at least 24h of data.